Extraction: model in config, chatter gate, geocode cache (server-26#298, server-26#104, server-26#88)

EXTRACTION_MODEL / EMBEDDING_MODEL replace the gpt-4o-mini and
text-embedding-3-small literals. gate_chatter (classify_chatter with roll_call
narrowed to whole-transcript roll calls) skips correction, extraction,
embedding and geocoding for housekeeping, writes skip_reason=non_event_chatter
and still runs the free clearance/plate backstops; upload.py's no-scene
fallback admits it like transcript_too_short. Ships OFF (CHATTER_GATE_ENABLED
=false): a gated call has no units, so it takes the no-fit-check thin attach,
and #127 requires a clean live shadow window first.
In-process TTL cache (24h, negatives 1h, 5000 entries, LRU) in front of the
Geocoding API, keyed by address+bounds; holds coordinates only, inside the
Maps 30-day lat/lng caching limit; errors and key/quota statuses not cached.
#104: the no-scene fallback passes the call doc's severity.
#88: resolved/reassignment parsed by _llm_bool, not bool().

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

drb-c2-core/app/config.py

@@ -79,8 +79,31 @@ class Settings(BaseSettings):
     # google-genai does not retry unless told to, so this is the whole bound.
     gemini_timeout_seconds: float = 60.0
 
+    # Scene extraction + embeddings (server-26#298). Were literals in
+    # intelligence.py; any OpenAI chat model that supports JSON mode works.
+    extraction_model: str = "gpt-4o-mini"
+    embedding_model: str = "text-embedding-3-small"
+    # Skip correction + extraction (+ embedding, geocoding) for transcripts
+    # chatter_classifier.gate_chatter calls radio housekeeping. OFF by default:
+    # a gated call carries no units, so it is thin and takes the no-fit-check
+    # thin attach (mechanism A) instead of the fit-tested path or the #115
+    # orphan gate, and server-26#127 set "0 false positives in a live shadow
+    # window" as the bar for going live. Read chatter_classifier_verdict off a
+    # shadow window first, then set CHATTER_GATE_ENABLED=true.
+    chatter_gate_enabled: bool = False
+
     # Google Maps (geocoding)
     google_maps_api_key: Optional[str] = None
+    # In-process cache of Geocoding API answers (server-26#298). Dispatch
+    # repeats the same address on every follow-up call of a job, and the
+    # place verifier re-geocodes the same nouns. In-process rather than
+    # Firestore: a Firestore read costs about what it saves, adds latency on
+    # every miss, and would persist Maps content -- the Maps terms allow
+    # caching lat/lng for at most 30 days, and an in-memory TTL can never
+    # outlive that. Lost on restart, which only costs a few repeat lookups.
+    geocode_cache_ttl_seconds: int = 24 * 60 * 60
+    geocode_cache_negative_ttl_seconds: int = 60 * 60  # ZERO_RESULTS etc.
+    geocode_cache_max_entries: int = 5000
     # 511NY developer key (server-26#183/#198). Issued with the NYSDOT Developer
     # Access Agreement grant; the keyless legacy API is gone since the 511x cutover.
     ny511_api_key: Optional[str] = None