feat: integrate memory retrieval and productivity workflows

[justinbuzzni]

Summary

• Integrate the remaining noble-koala work on top of the current main.
• Add evidence-focused memory retrieval, episode expansion, graduated evidence, and safer hook output behavior.
• Add memory field evaluation tooling, productivity health/reporting improvements, repository identity support, and related CLI/MCP workflows.
• Preserve the evidence-grounded usefulness and vector-writer hardening already merged through feat: add evidence-grounded usefulness and harden vector imports #33.

Branch analysis

• clever-hawk is fully contained in keen-tiger.
• keen-tiger and the current main have identical trees; that work was merged in feat: add evidence-grounded usefulness and harden vector imports #33.
• Only noble-koala still needed integration.
• This PR supersedes the conflicting feat: improve memory retrieval and productivity workflows #34 without modifying the original branch.

Why

The original noble-koala branch was based on the pre-#33 main, so its existing PR now conflicts with the usefulness dashboard and vector-writer concurrency changes. This integration reapplies the remaining feature work to the latest main and resolves those conflicts explicitly.

Conflict resolution

• Keep package version 1.0.59.
• Preserve the shared retrieval trace ID used by usefulness evidence while adopting question-focused evidence previews.
• Keep evaluation mode non-mutating for query traces and automatic graduation.
• Keep the long-lived semantic retrieval daemon read-only.
• Run scheduled graduation through a one-shot writer guarded by the existing project-scoped vector-worker.lock, preserving feat: add evidence-grounded usefulness and harden vector imports #33's LanceDB concurrency guarantees.

User and developer impact

• Higher-signal memory injection with traceable evidence and episode-aware recall.
• Automatic, bounded graduation without reintroducing concurrent vector writers.
• New field-evaluation and productivity-health workflows across CLI, server, and MCP surfaces.
• Repository identity and project registry behavior are more robust.

Validation

• npm run typecheck
• Focused integration tests: 6 files, 59 tests passed
• npm run test:run: 163 files, 970 tests passed
• npm run build
• npm run lint: 0 errors, 44 existing no-explicit-any warnings
• npm list claude-memory-layer: no self-dependency