feat(media-buy): add action rights assessment

[bokelley]

Summary

• add buyer assessment and routing helpers that join advisory product actions, accepted proposal change terms, and live MediaBuy availability without expanding authority
• add seller materialization/projection helpers with status, authorization, delegation, policy, condition, and portable-constraint gates
• restore beta.9 change-term runtime invariants in generated models and expose the relevant public types
• expand update decomposition, reference seller behavior, cross-SDK fixtures, and action-rights documentation

Compatibility

• preserves AdCP 3.1 terms_ref as opaque compatibility data
• projects pre-beta.9 and beta.9+ action shapes explicitly
• rejects mismatched aliases, invalid constraint/action pairs, duplicate terms, and unavailable or unknown routes
• treats seller_managed as normal asynchronous task processing

Verification

• uv run pytest -q — 7,430 passed, 77 skipped, 9 deselected, 1 xfailed
• make lint typecheck — Ruff clean; mypy clean across 1,235 source files
• 236 focused action, storyboard, codegen, public API, and response-builder tests
• released beta.9 read/control responses validated against the bundled schemas
• published JS storyboard runner against the reference seller — 90 selected scenarios passed

Closes #1067
Upstream: adcontextprotocol/adcp#6750
JavaScript parity: adcontextprotocol/adcp-client#2664

[aao-secretariat]

Ladon verdict: Approve

Approve — additive-only public surface with correct feat(media-buy) semver signal.

Checked:

• New media_buy_actions.py assessment/projection logic is fail-closed and keeps product/proposal/live surfaces separated as documented.
• Generated-model tightenings only bind beta.9-shaped payloads, so no existing-wire deserialization breaks (no forward-compat/discriminated-union regression).
• Type-system import layering preserved; no hand-edits treated as source outside legitimate regen. No CI-gate weakening. No credential storage in ctx_metadata.
• High-risk flag is true only on (modified) files, but the reviewer found no medium-or-higher concern tied to them, so the modifications are presumed contract-preserving.

Medium findings

• scripts/post_generate_fixes.py — enforce_change_term_runtime_constraints guards its validator insertions but not its import replace, so change_term_constraints.py ships a triple model_validator import and no longer matches a clean regen (posted inline, recorded).

Decision path: no critical/high findings; gated_paths false; high_risk reasons are all (modified) but only 1 medium finding exists (row 5 requires both — it fires only if a medium exists on a modified high-risk file; this medium is in a non-high-risk script under scripts/, so row 5 does not apply); no data-loss/schema/infra medium; no prior escalation; no no-auto-approve team; fewer than 3 mediums. Falls through to row 9 → approve.

[aao-secretariat]

Ladon verdict: Approve

Approve.

Clean pass. The delta is a well-tested idempotency fix for a prefix-collision bug in the codegen post-processing import rewrite (scripts/post_generate_fixes.py), its paired regenerated file collapsing a triple model_validator import to one, a regression test locking the fix, and a cosmetic Protocol stub change. No blocking or medium findings.

Gate checks: gated_paths is false, so row 2 does not apply despite review_decision: REVIEW_REQUIRED. high_risk is true, but all reasons are (modified) with no medium-or-higher findings, so rows 3–5 do not fire; the generated-file change is regeneration output paired with the codegen fix rather than a hand-edit. No no-auto-approve team match. Prior decision was also approve, so no sticky escalation. With zero findings, none of rows 1–8 fire — falls through to row 9.