Knowledge Refresh Synthesis

Update an existing synthesis when a fresh source lands in the base, without re-deriving it from scratch. This is the dominant compounding case: a new source supersedes or extends a prior synthesis, and you want to fold it in — adding to the prior evidence base, never replacing it.

The base already detects this: synthesis-impact.py (run at knowledge-ingest-source time) flags affected syntheses into binding.json::refresh_candidates[]. What was missing is a first-class path to act on a candidate. Doing it by hand means ~10 error-prone steps stitched across knowledge-ingest-source, knowledge-compose, knowledge-verify, and knowledge-finalize. The only existing one-shot route — knowledge-refresh push-mode — routes the topic into a full Phase-1 plan re-run whose knowledge-compose --source wiki re-derives the ingest-manifest via wiki-grounding, which under-maps (a real run mapped only 21 of 35 original sources), thinning the existing synthesis instead of extending it.