graphiti-core 0.29.0 builds FalkorSearchOperations as driver._search_ops in
FalkorDriver.__init__ but never assigns it to driver.search_interface.
search_utils.py dispatches on search_interface; without this one-line bridge
it falls back to interpreted-Cypher cosine math doing full table scans for
every entity dedup similarity check.
Combined with the vendored patches in graphiti_patches/ (restored in the
previous commit d2ec20e), this activates FalkorDB's native vector index for
the dedup similarity path. Empirical impact (per the original f645b74 commit
message): single-episode add_episode against a ~4,277-entity graph went from
indefinite hang to ~8.2 seconds.
Surgical restore: cherry-picks only the bridge code from f645b74 — not the
Pattern 1 async job model, not the v2.4 extraction instructions, neither of
which we want. Default extraction posture (taxonomy-naïve) stays the
operating mode. Rich-extraction story remains a BirdAI concern.