feat: add time_budget policy

[Yashika-1705]

Fixes #66

Adds the time_budget policy - a wall-clock ceiling per run. Trips at
observe when elapsed >= max_seconds and halts the run. Natural sibling
of step_cap for workflows whose cost is latency rather than tokens.

[tishachawla-jg]

Thanks @Yashika-1705 looks good overall. Tests pass locally. A few things:

• -observe calls view.window() every step, which copies the whole run history (see the note in Add a policy: the contributor path, and what is worth adding next #66 about window being O(N)). Since you only need the start time, you could store the first ts per run on the detector instead:

start = self._started.setdefault(attr.run_id, step.ts)

• ruff format --check fails on tests/test_policies_wrap_integration.py (missing blank line before the new test). make format fixes it.

• Docs: the first step doesn't hit an empty window (the ledger records the step before observe runs, so elapsed is 0.0), and the e2e row says max_seconds=0.0 but the test uses 0.5.

Please add a row to docs/product/policies-index.md.

Should be good to merge after that.