feat: учёт token usage в checks с агрегатами и UI

- миграция checks: nullable input_tokens/output_tokens/total_tokens/usage_json
  (идемпотентно, старые строки остаются NULL; probe мигрирует сам — cron стартует раньше uvicorn)
- probe.py и /api/check фиксируют usage ДО валидации контента: токены
  сохраняются при пустом/сломанном ответе; таймауты и ответы без usage = NULL, не нули
- usage_json хранит сырой usage провайдера (OpenAI/Anthropic/camelCase форматы)
- /api/checks и сегментная история несут токены; новый /api/token_stats:
  rolling 24h + all time, покрытие known/unknown, разбивка по провайдерам
  (провайдер из models с fallback на префикс model_id)
- UI: статы «Токены 24h» и «Покрытие usage» в шапке, токены в тултипах
  сегментов и логе проверок
- статика через __file__-относительный путь (работает и в /app/app, и локально)
- тесты (30, реальные HTTP против мока OmniRoute): миграция/идемпотентность,
  NULL при missing/zero usage, usage при упавшем контенте, таймаут, агрегаты,
  оба пути записи (probe.run + /api/check)

tests/test_api.py

@@ -0,0 +1,171 @@
+"""Ручной путь (main.py /api/check) + API-агрегаты token usage."""
+import json
+import sqlite3
+from datetime import datetime, timedelta, timezone
+
+
+def _con(db_path):
+    con = sqlite3.connect(db_path)
+    con.row_factory = sqlite3.Row
+    return con
+
+
+def _insert_check(db_path, model_id, at, ok=1, inp=None, out=None, tot=None, ujson=None):
+    con = _con(db_path)
+    con.execute(
+        "INSERT INTO checks (model_id, checked_at, ok, latency_ms, error, "
+        "input_tokens, output_tokens, total_tokens, usage_json) VALUES (?,?,?,?,?,?,?,?,?)",
+        (model_id, at, ok, 50, None, inp, out, tot, ujson),
+    )
+    con.commit()
+    con.close()
+
+
+def _iso(**kw):
+    return (datetime.now(timezone.utc) + timedelta(**kw)).isoformat()
+
+
+# ─── ручная проверка /api/check ───────────────────────────────────────────────
+
+def test_api_check_captures_usage(core, omni_server):
+    omni_server.set_completions(usage={"prompt_tokens": 12, "completion_tokens": 3, "total_tokens": 15})
+    r = core.client.get("/api/check/zai/test-model-a")
+    assert r.status_code == 200
+    body = r.json()
+    assert body["ok"] is True
+    assert body["input_tokens"] == 12
+    assert body["output_tokens"] == 3
+    assert body["total_tokens"] == 15
+
+    row = _con(core.db_path).execute(
+        "SELECT * FROM checks WHERE model_id='zai/test-model-a'").fetchone()
+    assert row["input_tokens"] == 12 and row["total_tokens"] == 15
+    assert json.loads(row["usage_json"])["prompt_tokens"] == 12
+
+
+def test_api_check_missing_usage_null(core, omni_server):
+    omni_server.set_completions(no_usage=True)
+    body = core.client.get("/api/check/zai/test-model-a").json()
+    assert body["ok"] is True
+    assert body["input_tokens"] is None and body["total_tokens"] is None
+
+    row = _con(core.db_path).execute(
+        "SELECT input_tokens, total_tokens, usage_json FROM checks WHERE model_id='zai/test-model-a'").fetchone()
+    assert row["input_tokens"] is None and row["total_tokens"] is None and row["usage_json"] is None
+
+
+def test_api_check_failed_content_preserves_usage(core, omni_server):
+    omni_server.set_completions(usage={"prompt_tokens": 8, "completion_tokens": 1, "total_tokens": 9},
+                                omit_choices=True)
+    body = core.client.get("/api/check/agy/test-model-b").json()
+    assert body["ok"] is False
+    assert body["input_tokens"] == 8  # usage пережил ошибку контента
+
+    row = _con(core.db_path).execute(
+        "SELECT ok, error, total_tokens FROM checks WHERE model_id='agy/test-model-b'").fetchone()
+    assert row["ok"] == 0 and row["error"]
+    assert row["total_tokens"] == 9
+
+
+def test_api_check_no_response_stores_null(core, omni_server):
+    # недоступный OmniRoute: ответа нет — usage остаётся NULL (не нулями)
+    saved = core.main.OMNIROUTE_URL
+    core.main.OMNIROUTE_URL = "http://127.0.0.1:1"  # connection refused, мгновенно
+    try:
+        body = core.client.get("/api/check/zai/test-model-a").json()
+    finally:
+        core.main.OMNIROUTE_URL = saved
+    assert body["ok"] is False
+    assert body["input_tokens"] is None and body["total_tokens"] is None
+
+    row = _con(core.db_path).execute(
+        "SELECT ok, error, total_tokens, usage_json FROM checks WHERE model_id='zai/test-model-a'").fetchone()
+    assert row["ok"] == 0 and row["error"]
+    assert row["total_tokens"] is None and row["usage_json"] is None
+
+
+# ─── истории: usage в /api/checks и /api/models.history ──────────────────────
+
+def test_api_checks_history_includes_usage(core, omni_server):
+    omni_server.set_completions(usage={"prompt_tokens": 5, "completion_tokens": 2, "total_tokens": 7})
+    core.client.get("/api/check/zai/test-model-a")
+    omni_server.set_completions(no_usage=True)
+    core.client.get("/api/check/zai/test-model-a")
+
+    data = core.client.get("/api/checks/zai/test-model-a").json()
+    assert len(data["checks"]) == 2
+    with_usage = [c for c in data["checks"] if c["total_tokens"] == 7]
+    without = [c for c in data["checks"] if c["total_tokens"] is None]
+    assert len(with_usage) == 1 and with_usage[0]["input_tokens"] == 5
+    assert with_usage[0]["usage_json"]
+    assert len(without) == 1 and without[0]["input_tokens"] is None
+
+    # сегментная история тоже несёт токены (для тултипов UI)
+    models = core.client.get("/api/models").json()
+    model = next(m for m in models if m["id"] == "zai/test-model-a")
+    toks = [h["total_tokens"] for h in model["history"]]
+    assert set(toks) == {7, None}
+
+
+# ─── /api/token_stats ─────────────────────────────────────────────────────────
+
+def test_token_stats_empty_db(core):
+    ts = core.client.get("/api/token_stats").json()
+    assert ts["window_24h"]["checks"] == 0
+    assert ts["window_24h"]["coverage_pct"] is None
+    assert ts["all_time"]["checks"] == 0
+    assert ts["providers"] == []
+
+
+def test_token_stats_windows_coverage_providers(core):
+    # zai: 2 свежих с usage + 1 старая (>24h) с usage
+    _insert_check(core.db_path, "zai/test-model-a", _iso(minutes=-10), inp=10, out=5, tot=15)
+    _insert_check(core.db_path, "zai/test-model-a", _iso(minutes=-20), inp=20, out=2, tot=22)
+    _insert_check(core.db_path, "zai/test-model-a", _iso(hours=-30), inp=100, out=50, tot=150)
+    # agy: 1 свежая без usage + 1 старая с usage
+    _insert_check(core.db_path, "agy/test-model-b", _iso(minutes=-5))
+    _insert_check(core.db_path, "agy/test-model-b", _iso(hours=-40), inp=7, out=3, tot=10)
+    # модель, которой нет в таблице models: провайдер выводится из model_id
+    _insert_check(core.db_path, "orphan/x", _iso(minutes=-1), inp=5, out=5, tot=10)
+
+    ts = core.client.get("/api/token_stats").json()
+
+    w = ts["window_24h"]
+    assert w["checks"] == 4 and w["known"] == 3 and w["unknown"] == 1
+    assert w["coverage_pct"] == 75
+    assert (w["input_tokens"], w["output_tokens"], w["total_tokens"]) == (35, 12, 47)
+
+    a = ts["all_time"]
+    assert a["checks"] == 6 and a["known"] == 5 and a["unknown"] == 1
+    assert (a["input_tokens"], a["output_tokens"], a["total_tokens"]) == (142, 65, 207)
+
+    provs = {p["provider"]: p for p in ts["providers"]}
+    assert set(provs) == {"zai", "agy", "orphan"}
+    assert provs["zai"]["window_24h"]["total_tokens"] == 37
+    assert provs["zai"]["all_time"]["total_tokens"] == 187
+    assert provs["agy"]["window_24h"]["known"] == 0
+    assert provs["agy"]["window_24h"]["unknown"] == 1
+    assert provs["agy"]["window_24h"]["total_tokens"] is None  # сумма по NULL-группе
+    assert provs["agy"]["all_time"]["total_tokens"] == 10
+    assert provs["orphan"]["window_24h"]["total_tokens"] == 10  # fallback из model_id
+
+
+def test_token_stats_counts_both_paths(core, omni_server):
+    """probe.run() и ручной /api/check пишут usage, агрегат считает оба пути."""
+    omni_server.set_completions(usage={"prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15})
+    core.probe.run()  # путь cron: zai 15 + agy 15
+    core.client.get("/api/check/zai/test-model-a")  # путь ручной: +15
+    omni_server.set_completions(no_usage=True)
+    core.client.get("/api/check/agy/test-model-b")  # путь ручной без usage: NULL
+
+    ts = core.client.get("/api/token_stats").json()
+    w = ts["window_24h"]
+    assert w["checks"] == 4          # 2 probe + 2 ручных
+    assert w["known"] == 3           # probe×2 + ручной zai
+    assert w["unknown"] == 1         # ручной agy без usage
+    assert w["total_tokens"] == 45   # 15 × 3 известных
+    assert w["input_tokens"] == 30 and w["output_tokens"] == 15
+
+    provs = {p["provider"]: p for p in ts["providers"]}
+    assert provs["zai"]["window_24h"]["total_tokens"] == 30      # probe + manual
+    assert provs["agy"]["window_24h"]["total_tokens"] == 15       # только probe