Optimize CogMap extraction orchestration

[ryannadel]

Large CogMap refreshes currently rely on the host agent to infer efficient batching, concurrency, validation, and resume behavior. This makes high-volume runs slower and leaves malformed or retried extraction outputs vulnerable to silent acceptance or duplicate ingestion.

Approach

• adapt extraction sizing toward roughly 8-10 batches for large deltas, bounded by character and chunk limits, with a maximum of eight concurrent extraction workers
• emit pending-only batch work, progress, diagnostics, and machine-readable concurrency guidance in action.json
• validate UTF-8, JSON, the exact top-level schema, required fields and types, enums, concept references, and batch-local provenance before ingestion
• retain valid partial outputs across retries and archive batches under deterministic IDs, including migration of matching legacy timestamp archives
• split pathological oversized source chunks before batching and enforce sequential extraction, resolution, and synthesis gates
• update the canonical and mirrored skill instructions so hosts follow the optimized loop reliably

Validation

• python -m compileall -q .
• python -m unittest discover -s tests -v (9 tests)
• python cogmap-app/tools/sync_skill.py --check
• GitHub Actions CI