refactor(mvp): harden runtime and streaming flow

[talhasarit]

Summary

• add lifespan-managed app settings/runtime resources, explicit store selection, and honest health/readiness semantics
• harden adapters, pipeline state transitions, and SSE replay/error handling for a reliable PRD-only MVP
• narrow the frontend/docs/package surface to implemented behavior and add regression coverage for runtime, streaming, pipeline, frontend helpers, and CLI wiring

Verification

• ruff check .
• ruff format --check .
• mypy backend frontend tests
• pytest -q

---

Note

Medium Risk
Touches core request/streaming flow, runtime lifecycle, and LLM adapter error handling; regressions could break PRD runs or SSE replay despite added test coverage.

Overview
Refactors the backend into a lifespan-managed runtime with centralized AppSettings, structured logging, and explicit state-backend selection (redis/memory/auto fallback) plus a new /ready endpoint that reflects store health.

Hardens generation/streaming by persisting richer PRDState (adds idea + terminal error), broadcasting via a shared fan-out StreamerService, and making /api/v1/stream/{run_id} replay the latest persisted state before streaming subsequent revisions (including terminal completion/error states).

Updates vanilla LLM adapters to use typed AdapterError, configurable models/timeouts, the google-genai SDK, and structured success/failure logging; pipeline execution now records truthful step transitions and persists terminal errors without faking successful content. Frontend/docs/env/Docker packaging are narrowed to shipped PRD-only behavior, and extensive unit/integration tests are added for runtime selection, SSE replay, pipeline transitions, CLI wiring, and frontend helpers.

^{Written by Cursor Bugbot for commit bb46c14. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}