docs: document zero-threshold policy fix and improve ledger/policy inline docs

[cursor]

Docs added/updated

CHANGELOG.md

Added a ### Fixed entry under Unreleased for the zero-threshold policy sample bug merged in PR #3. The entry explains what broke (falsy or-based fallback treated an explicit 0 like None), what was fixed (is not None guards), and what it enables (a policy can set thresholds to 0 to unconditionally accept any sample size).

src/flightdeck/ledger.py

Added docstrings to three previously undocumented functions:

• confidence_label — documents the HIGH / MEDIUM / LOW tier logic, explains that a threshold of 0 is valid and means "no minimum required", and notes the empty-window case.
• evaluate_policy — lists each constraint checked (max_cost_per_run_usd, max_latency_ms, max_error_rate, require_high_diff_confidence) and documents the skip conditions.
• diff_releases — full Args/Returns/Raises docstring covering: threshold resolution (None → config default, 0 → explicit override), the agent-id invariant for cross-agent diff rejection, how window is used (label only, not a filter), and the returned DiffResult.

src/flightdeck/models.py

Added a class docstring to Policy that documents the two field groups and their distinct semantics:

• Constraint fields (max_*) — None disables the constraint.
• Sample threshold fields (min_*) — None defers to workspace config; 0 unconditionally accepts any sample size (the key distinction introduced by PR fix: respect zero policy sample thresholds #3).

schemas/v1/policy.schema.json

Regenerated (via scripts/generate_schemas.py) to include the new Policy docstring as the JSON Schema description field. All 49 tests pass including test_committed_json_schemas_match_models.

Which codepaths these cover

• flightdeck.ledger.diff_releases — the primary comparison engine (cost, latency, error-rate rollups, confidence labelling, policy evaluation).
• flightdeck.models.Policy — the on-disk and in-memory policy object consumed by diff_releases and the CLI's release diff / release promote / release rollback commands.

Key knowledge gaps addressed

1. The None vs 0 threshold contract was implicit in the code and absent from any docs; it is now explicit in both Policy's docstring and diff_releases's docstring.
2. What confidence_label thresholds actually mean (including the 0-is-valid edge case) was previously only discoverable by reading the implementation.
3. Which constraints evaluate_policy checks and when it skips them was undocumented.
4. The bug and its fix are now captured in CHANGELOG.md so future readers understand why is not None is used instead of or.