feat(adcp): add force_create_media_buy_arm + force_task_completion controller scenarios

[bokelley]

Closes #281

Adds two new TestControllerStore scenarios for storyboard parity with the Node training-agent (adcp#3115, adcp#3194). Sellers running the create_media_buy_async.yaml storyboard suite against a Python reference seller will now grade passing rather than not_applicable on the submitted-arm phase.

Changes

• src/adcp/server/test_controller.py — adds force_create_media_buy_arm and force_task_completion to SCENARIOS, adds abstract base stubs with full docstrings (including idempotency/isolation contract examples), adds dispatcher branches with centralized input validation (arm enum, conditional task_id-when-submitted, char limits, non-empty result dict, whitespace-strip on task_id for both scenarios), and derives register_test_controller's inline enum from SCENARIOS so it can't drift.
• src/adcp/server/mcp_tools.py — adds both new scenario names to ADCP_TOOL_DEFINITIONS comply_test_controller enum (the static stub; the Pydantic-generated path also now carries them).
• src/adcp/types/generated_poc/compliance/comply_test_controller_request.py — adds force_create_media_buy_arm and force_task_completion to the Scenario enum and arm/task_id/result fields to Params, so _apply_pydantic_schemas() generates the correct live MCP tool schema rather than clobbering the new names with the stale generated enum.
• schemas/cache/compliance/comply-test-controller-request.json — adds both scenario names to the request schema enum, arm/task_id/result param properties, and allOf conditional constraints.
• schemas/cache/compliance/comply-test-controller-response.json — adds ForcedDirectiveSuccess as a new oneOf variant (required: success, arm; discrimination via not: required:[previous_state|scenarios|simulated|error]), and adds both new names to ListScenariosSuccess.scenarios.items.enum.
• tests/test_force_create_media_buy_arm_and_force_task_completion.py — 19 new tests at parity with the Node training-agent nine-test pattern: valid registration (submitted + input-required arms), overwrite-pending, INVALID_PARAMS for missing/bad arm, task_id conditional requirement, char limits, whitespace task_id rejection, force_task_completion valid/idempotent/diverging/cross-account, INVALID_PARAMS for missing task_id/empty result/long task_id, list_scenarios advertisement (both implemented, neither implemented, partial).

What was tested

• pytest tests/ (excluding slow integration + ip_pinned_transport): 2181 passed, 21 skipped, 0 failures
• ruff check src/: clean
• mypy src/adcp/server/test_controller.py: no errors
• Verified ADCP_TOOL_DEFINITIONS["comply_test_controller"] enum at import time includes both new names after the Pydantic schema generation pipeline runs

Pre-PR review

• code-reviewer: approved — all blockers addressed (Pydantic enum clobber fixed via generated model update; whitespace task_id stripped in both dispatchers); 4 nits noted below
• ad-tech-protocol-expert: approved — ForcedDirectiveSuccess oneOf discrimination is airtight across all four sibling variants; NOT_FOUND cross-account isolation matches AdCP caller-isolation semantics; both original blockers resolved; one nit (whitespace task_id also absent from force_task_completion) fixed in follow-up commit

Nits surfaced (not fixed — follow-up candidates)

• arm="" produces "Missing required parameter: 'arm'" rather than the more accurate "arm must be 'submitted' or 'input-required'" — misleading wording, correct error code
• Non-string task_id (e.g. integer from JSON) passes the not-check and hits len() with TypeError, landing in INTERNAL_ERROR instead of INVALID_PARAMS — pre-existing pattern in other dispatchers; low-severity since MCP clients are typed
• _BothStore in tests has a redundant explicit TestControllerStore base (already inherited via both parents)
• Test store overrides use **kwargs rather than the documented *, context=None opt-in pattern

---

Triage-managed PR. This bot does not currently iterate on
review comments or PR conversation threads (only on the source
issue). To unblock:

• Push fixup commits directly: gh pr checkout <num> →
  fix → push.
• Or re-trigger: comment /triage execute on the source
  issue.

See adcp#3121
for context.

Session: https://claude.ai/code/session_01HcWEyL1xWQtFTmDEaMme8Q

---

Generated by Claude Code