docs(skills): schema refresh + brand-rights governance flow

[bokelley]

Summary

Follow-up to PR #709. The npm run compliance:skill-matrix dogfood run (14 skill × storyboard pairs) had 0/14 pass — but most failures stemmed from skill examples showing stale schemas that upstream redefined in the last several days (#691–#708). This PR refreshes the skill examples and closes one real coverage gap in brand-rights.

Root-cause fixes

1. list_creative_formats.renders[] (~6 pairs)
   Upstream redefined renders to require role plus oneOf { dimensions: object } or { parameters_from_format_id: true }. Skills showed renders: [{ width, height }] which fails validation. Updated seller, creative, generative-seller, retail-media examples and flagged the old shorthand explicitly.

2. get_media_buys.media_buys[].currency + .total_budget (~4 pairs)
   Now required per row. Seller skill example shows both + a persistence note: save these on create_media_buy so subsequent reads can echo them.

3. context response field shape (~4 pairs, signals/si/governance)
   Schema-typed as object. All 8 skills previously recommended context: args.context echo, which fabricates strings when args.context is undefined or when Claude conflates it with domain fields like campaign_context. New guidance: don't set context at all — createAdcpServer auto-injects the request's context object post-handler; hand-setting a non-object string fails validation and the framework does not overwrite.

4. Brand-rights governance_denied flow (1 pair)
   Scenario expects check_governance call before acquire_rights. Added accounts.syncAccounts + accounts.syncGovernance handlers to the implementation example, plus a checkGovernance() call in acquireRights that returns GOVERNANCE_DENIED with findings propagated from the buyer's governance agent.

5. Seller idempotency footnote (1 pair)
   Referenced Storyboard runner: raw-HTTP probe for idempotency missing-key vector (bypasses SDK auto-injection) #678 as a known grader-side MCP Accept-header issue so builders don't chase a skill fix for what's a grader bug.

Expected matrix delta

Baseline: 0/14 passed. Conservative estimate with these fixes: 10–12/14. Remaining failures likely need case-by-case diagnosis (e.g., pending_creatives → pending_start transition, creative state-machine quirks).

Test plan

• All 8 skills compile clean in isolation (markdown only — no code)
• Re-run npm run compliance:skill-matrix on this branch once merged
• Validate remaining failures are new signatures, not regressions of what this PR fixed

🤖 Generated with Claude Code