feat(examples): handle @adcp/sdk@7.10.x storyboard scenarios (closes #782)

[bokelley]

Summary

Bumps ADCP_SDK_VERSION in .github/workflows/ci.yml from 7.9.0 to 7.10.2. Closes #782.

The two failing scenarios called out in the issue both pass against examples/seller_agent.py on @adcp/sdk@7.10.2 with no example-side changes:

• media_buy_seller/measurement_terms_rejected/create_media_buy_aggressive_terms — the scenario uses $context.product_id from get_products. seller_agent.py's get_products returns the static PRODUCTS catalog starting with premium-homepage, which valid_ids already contains. The existing TERMS_REJECTED guard (variance<5 or window not in c3/c7) fires correctly on the aggressive payload (max_variance_percent: 0, window: c28) before any PRODUCT_NOT_FOUND fall-through. The seller's error-code precedence is already correct for the 7.10.x scenario shape.

• media_buy_seller/invalid_transitions/update_unknown_package — the scenario's preceding setup phase creates a real media buy and stores it in $context.media_buy_id. The update_unknown_package step uses that real id with a fabricated package_id, so the seller's existing media-buy-then-package precedence ordering returns PACKAGE_NOT_FOUND correctly. No reordering needed.

Why the 7.10.1 failures didn't reproduce on 7.10.2

The original failures in the #782 report were observed against @adcp/sdk@7.10.1, which shipped a packaging regression (v1-canonical-mapping.json lived at dist/lib/schemas-data/... instead of the path the runtime looked up at schemas/cache/...). CI worked around it by copying tests/fixtures/v1-canonical-mapping.json into the SDK install — but that fixture was missing the new 7.10.x scenario schema entries, so the runner fell through to the wrong validators and surfaced misleading TERMS_REJECTED → PRODUCT_NOT_FOUND and PACKAGE_NOT_FOUND → MEDIA_BUY_NOT_FOUND failures. 7.10.2 fixes the lookup path; with the canonical mapping resolving from the SDK's own ship, both scenarios now validate against the correct schemas and the example passes unchanged.

Test plan

• examples/seller_agent.py storyboard suite passes 57/57 graded steps against @adcp/sdk@7.10.2 locally (2 unchanged list_creatives missing-tool skips).
• Existing pytest tests/ passes (4938 passed, 34 skipped, 1 xfailed).
• CI confirms the bump green across all 4 storyboard jobs (examples/seller_agent.py, multi_platform_seller, sales-proposal-mode, v3 reference seller (translator)).
• No new failures vs. 7.9.0 baseline on the other three storyboard runners (the multi-platform seller has 5 pre-existing failures on both 7.9.0 and 7.10.2 that are not part of this PR's scope).

🤖 Generated with Claude Code

[aao-ipr-bot]

LGTM. Follow-ups noted below.

Single-line bump of ADCP_SDK_VERSION from 7.9.0 to 7.10.2. PR description and commit body lay out exactly why no example-side change was needed — 7.10.1 was the packaging regression (canonical mapping at dist/lib/schemas-data/... vs. runtime lookup at schemas/cache/...), 7.10.2 fixes the lookup path, and examples/seller_agent.py's existing error-code precedence (TERMS_REJECTED guard, media-buy-then-package ordering) already lines up with the 7.10.x scenario shapes.

Things I checked

• Diff is exactly the one-line bump at .github/workflows/ci.yml:12. No collateral changes.
• Cache invalidation note in the workflow comment at .github/workflows/ci.yml:10-11 is accurate — the env var is the cache key, so bumping it busts the install cache on the next run.
• Skip-everything list applies: pure dependency bump, no src/adcp/ touched, no public API surface, no generated types. No experts called.
• Local validation in the PR body is concrete and quantified: 57/57 graded steps on seller_agent.py, 4938 passed, 34 skipped, 1 xfailed on pytest tests/.
• The two unchecked test-plan items (CI confirms green across all 4 storyboard jobs, no new failures vs. 7.9.0 baseline) are CI-gated, not manual — author already noted the pre-existing multi-platform-seller failures are out of scope. Not a manual-verification gap.

Follow-ups (non-blocking — file as issues)

• Conventional-commit prefix triggers an unwanted minor bump. The commit is feat(examples): but the diff is purely .github/workflows/ci.yml. Per release-please-config.json (bump-minor-pre-major: true), this will bump adcp's minor version for a CI-only change. ci: or chore(ci): would have been the right shape — neither triggers a version bump pre-major. Worth a follow-up on the commit-prefix convention for CI-pinned-dep bumps; not blocking this PR since the semver signal isn't a breaking-change miss, just a noisier bump than intended.

LGTM.