feat: memory pipeline resilience — retry, queue, and stale observation cleanup

[not-reed]

• Add exponential-backoff retry to embedding/observer/reflector API calls (handles 5xx, 429 with Retry-After, network errors)
• Introduce PipelineQueue for crash-recoverable post-turn memory processing; jobs are persisted, stuck "running" jobs reset on startup, dead-letter after 3 attempts
• Parallelize embedding generation in promoteObservations; wrap supersede + insert in transactions; mark promoted_at per-observation so partial progress survives crashes
• Make WorkerModelConfig.baseUrl required and wire OPENROUTER_BASE_URL env across construct/cortex/loom
• Add memory health action surfacing queue + memory stats + recent dead-letter failures
• Migration 014: pipeline_jobs table
• Migration 015: expire active observations with unresolved "this/next/past weekend" references whose date has passed
• Update observer/reflector prompts to resolve weekend phrases at observation time and never collapse resolved dates back to relative form
• (carries previously-uncommitted) Consolidate 28 tools into 9 action-based tools; decompose processMessage into 4 subroutines

[cloudflare-workers-and-pages]

Deploying sprawl with    Cloudflare Pages

Latest commit:	c347a06
Status:	✅  Deploy successful!
Preview URL:	https://a70772d6.sprawl.pages.dev
Branch Preview URL:	https://not-reed-memory-pipeline-res.sprawl.pages.dev

View logs