feat(decisioning): ProposalCapabilities.auto_commit_on_put_draft (#723)

[bokelley]

Closes #723 (option B from the issue).

Summary

Pre-#723 the storyboard step media_buy_seller/proposal_finalize/create_media_buy was unreachable for managers that didn't declare finalize=True — brief mode returned proposals as DRAFT, next call's try_reserve_consumption raised PROPOSAL_NOT_COMMITTED. Adopters wrote state=COMMITTED directly inside their put_draft implementations (salesagent PR #390); this PR ships the framework-side equivalent.

Surface

class ProposalCapabilities:
    auto_commit_on_put_draft: bool = False           # NEW
    auto_commit_ttl_seconds: int = 604800            # NEW — 7-day default

When auto_commit_on_put_draft=True, the framework calls ProposalStore.commit immediately after put_draft on every proposal returned, promoting DRAFT → COMMITTED in a single dispatch.

Validation

• auto_commit_on_put_draft=True + finalize=True → AdcpError. Mutually exclusive — both on would race.
• auto_commit_ttl_seconds <= 0 → AdcpError. Zero would expire on commit.

Why option B (not A, C, D from the issue)

• B is the smallest contained framework change. Adopters opt in per-manager; existing managers see no behavior change.
• A (storyboard finalize step) is spec-honest but heavy: every adopter ships finalize_proposal.
• C (split storyboards) is a spec-repo conversation, not a framework change.
• D (docs only) doesn't close the gap.

Adopter migration

Salesagent drops the inline state=COMMITTED in SalesAgentProposalStore.put_draft and sets auto_commit_on_put_draft=True on its ProposalCapabilities.

Test plan

• 7 new tests covering: happy path; back-compat default; TTL honored; multi-proposal; mutually-exclusive guard; ttl<=0 guard; default-False back-compat.
• All 84 proposal-lifecycle tests pass locally.
• ruff + mypy clean.

🤖 Generated with Claude Code