[WIP] Currently idea decomposition (IdeaDecomposer) runs as a synchronous LLM

[bamdadd]

⚠️ This is a draft PR created by Lintel after a pipeline failure.
The automated workflow could not complete successfully. An engineer should review the changes, fix any issues, and mark the PR as ready.

Summary

Implement a new domain feature following the established lintel monorepo patterns: define contracts first, implement persistence and API packages, mount in the app composition root, and validate with unit and integration tests.

Changes

• Audit existing codebase structure and document gaps
• Define domain contracts: types, commands, and events in packages/contracts/
• Implement domain persistence layer in packages/persistence/ or a new domain package
• Create the domain API package under packages/-api/
• Mount new API router in packages/app/ composition root
• Write unit tests for contracts and domain logic
• Write integration and Postgres-backed tests
• Lint, typecheck, and validate full pipeline

Context

Currently idea decomposition (IdeaDecomposer) runs as a synchronous LLM call inside the chat service handler. This gives zero visibility — no pipeline stages, no SSE progress events, no retry capability. Convert it to a proper workflow (e.g. idea_to_work_items) with stages: classify_idea → decompose → validate_coverage → create_work_items. This gives: pipeline visibility in UI, SSE events for progress, audit trail, and ability to retry. The chat intent handler should create a pipeline run and return immediately with a link, rather than blocking for 60+ seconds. Package: packages/chat-api and packages/workflows. Out of scope: changing the decomposer prompt itself.

---

Raised by Lintel 🤖

[bamdadd]

Closing stub PR from failed pipeline. Will retry with ingest fix.