feat(clio): auto-ingest documenter output (docs/**/*.md) into Clio

[fstamatelopoulos]

Summary

Closes a long-standing inconsistency: cfcf auto-ingests almost every workspace artifact into Clio (iteration logs, judge assessments, reflection analyses, plan.md, decision-log, architect-review, problem-pack, context-pack…), but explicitly excluded the documenter's docs/*.md output. The exclusion's stated rationale ("docs/ is canonical, Clio is redundant") applied equally to plan.md — which IS auto-ingested. The carve-out was inconsistent and cost cross-workspace discoverability of the most polished, integrative artifact a workspace produces.

Surfaced by user question while planning v0.24.4 dogfood of the freshly-finished gmbot run: "Is the documenter ingesting the generated docs into Clio? Or better the harness doing that?" — neither, by design. The design was wrong. This PR fixes it.

Behaviour

After the documenter completes (both auto-document path inside the iteration loop AND standalone cfcf document):

• Walk <repo>/docs/ recursively
• For each *.md file: ingest into Clio as a separate document
• Stable per-file title: <workspace>: docs/<relative-path> (e.g. gmbot: docs/architecture.md, gmbot: docs/api/auth.md)
• updateIfExists: true — re-running the documenter overwrites in place, never duplicates; sha256 dedup makes unchanged content a no-op
• Author stamp: documenter|<adapter>|<model>
• Metadata: {role: \"documenter\", artifact_type: \"documenter-output\", file_path: \"<rel>\", tier: \"semantic\", ingest_trigger: \"loop-auto\" | \"manual\", …} — the documenter-output artifact_type makes the new docs filterable via cfcf clio search --metadata
• Non-.md files ignored (images, JSON config); dot-dirs (.git, .vscode) skipped; empty files skipped
• Per-file errors logged + counted but never fail the batch (best-effort, same policy as other auto-ingest hooks)
• Respects clio.ingestPolicy: \"off\" → no-op; \"summaries-only\" and \"all\" → runs

Files

• packages/core/src/clio/loop-ingest.ts — new ingestDocumenterOutput() + walkMarkdownFiles() helper
• packages/core/src/iteration-loop.ts — call site inside auto-document branch
• packages/core/src/documenter-runner.ts — call site for standalone cfcf document
• packages/core/src/templates/cfcf-documenter-instructions.md — prose updated to match reality

Test plan

• bun run test — 1049 tests pass (+11 new)
• bun run typecheck — clean
• 11 new tests in loop-ingest.test.ts cover: multi-file ingest with stable titles, recursive walk, updateIfExists round-trip (one doc per file, content updates in place), non-.md filtering, dot-dir skipping, missing docs/ no-op, empty file skipping, author stamp, trigger metadata (loop-auto vs manual), clio.ingestPolicy: \"off\" → no-op, clio.ingestPolicy: \"summaries-only\" → runs
• Manual smoke (post-merge): run cfcf document on a workspace with docs/architecture.md; verify cfcf clio search \"<term>\" --metadata '{\"artifact_type\":\"documenter-output\"}' returns the doc with the expected title format

Versioning

Target: v0.24.4 patch. New behaviour is additive and gated by the existing clio.ingestPolicy (no opt-out is needed beyond the existing config knob). Same shape as v0.24.3's release.

🤖 Generated with Claude Code