feat: export cost and governance data as CSV/JSONFeat/csv json export token governance

[RjyavardhanSingh]

What

targets issue #145
On-demand export of run-records and governance events from the control plane's
SQLite store as CSV (finance/spreadsheets) and JSON (pipelines). Two surfaces:
an GET /v1/export HTTP endpoint and a Streamlit Dashboard "Export" expander.

Why

The plane holds all cost and governance data but there was no way to pull it into
a spreadsheet, a cost tool, or an archive. This is the FinOps view (who spent what)
and the compliance audit trail (halts, MUTATE/INJECT decisions, which policy fired).

Changes

Backend — src/tokenops/control/store.py

• Added _coerce_epoch() helper: defensively converts epoch floats, ISO date
  strings, datetime.date, and datetime objects to epoch floats. Prevents the
  SQLite type-mismatch bug where REAL >= TEXT silently returns zero rows.
• Added Store.export_runs(): filtered query over the runs table supporting
  from_at, to_at, agent, status, tenant, and limit (capped at 10 000).
  Ordered by started_at DESC.

Backend — src/tokenops/control/http.py

• Mounted GET /v1/export on the FastAPI app.
• Query params: from_at, to_at, agent, status, tenant, format (json|csv),
  limit.
• CSV columns: run_id, agent, status, cost_micros, steps, started_at,
  ended_at, duration_s, dims, halt_reason, detector, governance_events.
• JSON response includes X-Total-Count header; CSV includes
  Content-Disposition: attachment.

Backend — src/tokenops/server/app.py

• Imported and mounted mount_export(app, store).

UI — src/tokenops/ui/views/dashboard.py

• Added "Export run data" expander with four filter controls (From, To, Agent, Status).
• Displays "N runs matched" count.
• Download buttons for CSV and JSON.
• _date_to_epoch() converts Streamlit date_input values to epoch floats
  (start-of-day for from_at, end-of-day for to_at).

Tests — tests/test_export.py

• 25 tests: Store.export_runs() unit tests (filters, limit, empty, governance
  events), GET /v1/export integration tests (JSON, CSV, empty, limit cap),
  _coerce_epoch() unit tests (None, float, int, date string, datetime string,
  date object, datetime object), and regression tests for string-date filtering.

Also in this branch

• src/tokenops/control/pricing.py: added Gemini model rates.
• src/tokenops/providers/openai.py: added OPENAI_BASE_URL passthrough.

Acceptance criteria

• Pull run-records + governance events over a time range in CSV and JSON.
• Scoped by read auth and tenant; no cross-tenant export (tenant filter).
• Admin "Export" button in Dashboard.
• No DB schema changes — new query only, existing tables untouched.
• No breaking changes to existing API — new GET /v1/export route only.
• All 283 tests pass, 0 failures.
• Lint and format clean.

How to test

1. make install && make run (plane on :7700, dashboard on :8501)
2. make demo (agents register runs in tokenops.db)
3. Open localhost:8501 → scroll to "Export run data" → pick filters → Download CSV
4. Or via API: curl http://localhost:7700/v1/export?format=csv

Results

CSV

JSON

[
  {
    "run_id": "run-5",
    "agent": "researcher",
    "status": "error",
    "parent_run": null,
    "parent_span": null,
    "halt_reason": null,
    "detector": null,
    "cost_micros": 12000,
    "steps": 6,
    "started_at": 1789762067.1769028,
    "ended_at": 1789762077.1769028,
    "task": null,
    "dims": {
      "tenant": "globex"
    },
    "governance_events": [
      {
        "kind": "halt",
        "reason": "model not found",
        "policy": "pre_call_worst_case"
      }
    ]
  },
  {
    "run_id": "run-4",
    "agent": "editor",
    "status": "completed",
    "parent_run": null,
    "parent_span": null,
    "halt_reason": null,
    "detector": null,
    "cost_micros": 3000,
    "steps": 3,
    "started_at": 1789761767.1769006,
    "ended_at": 1789761787.1769009,
    "task": null,
    "dims": {
      "tenant": "acme"
    },
    "governance_events": []
  },
  {
    "run_id": "run-3",
    "agent": "writer",
    "status": "completed",
    "parent_run": null,
    "parent_span": null,
    "halt_reason": null,
    "detector": null,
    "cost_micros": 8000,
    "steps": 5,
    "started_at": 1789760867.1768982,
    "ended_at": 1789760917.1768985,
    "task": null,
    "dims": {
      "tenant": "globex"
    },
    "governance_events": []
  },
  {
    "run_id": "run-1",
    "agent": "researcher",
    "status": "completed",
    "parent_run": null,
    "parent_span": null,
    "halt_reason": null,
    "detector": null,
    "cost_micros": 15000,
    "steps": 12,
    "started_at": 1789759067.1768858,
    "ended_at": 1789759167.1768873,
    "task": null,
    "dims": {
      "tenant": "acme"
    },
    "governance_events": [
      {
        "kind": "mutate",
        "reason": "cap applied",
        "policy": "pre_call_worst_case"
      }
    ]
  },
  {
    "run_id": "run-2",
    "agent": "researcher",
    "status": "halted",
    "parent_run": null,
    "parent_span": null,
    "halt_reason": null,
    "detector": null,
    "cost_micros": 50000,
    "steps": 8,
    "started_at": 1789755467.1768959,
    "ended_at": 1789755567.1768959,
    "task": null,
    "dims": {
      "tenant": "acme"
    },
    "governance_events": [
      {
        "kind": "halt",
        "reason": "budget exceeded",
        "policy": "cost_budget"
      }
    ]
  }
]