v1.1.1

Fixes

• Pool crash resilience: The worker pool now survives transient I/O errors (e.g. disk full) without dying. A supervisor wrapper (_resilient_pool) automatically restarts the pool on unexpected crashes, with a circuit breaker to prevent infinite restart loops.
• Error isolation: _snapshot() and _handle_result() OSErrors are caught individually instead of killing the scheduling loop.

Testing

• 11 new E2E pipeline tests using a ClaudeSDKClient digital twin — exercises the full pipeline (idea submission → pool scheduling → agent dispatch → phase transitions → release) without requiring Claude API auth.
• tests/sdk_twin.py: Drop-in FakeClaudeSDKClient replacement that honors the same API contract (AssistantMessage, ToolUseBlock, ResultMessage), enabling CI testing without API keys.
• Tests cover: full pipeline traversal, context passing, cost tracking, multi-idea scheduling, crash recovery, phase history integrity.

CI

• All E2E tests run in GitHub Actions via the existing e2e pytest marker — no Claude auth required.