hermes(stt): greedy single-temperature final decode for fast startup

Beam-2 with a temperature-fallback ladder made the final-model warmup
run all three temperature retries under beam search before the server
bound its port, so /health was refused for ~8 min and STT was down that
whole time on every roll (and hinted at slow per-utterance decodes).
Production now runs the accurate large-v3-turbo model greedily at a
single temperature, keeping the proper-noun priming prompt that fixes
names like Amy/Córdoba - fast startup, fast decodes, accuracy intact.
Beam stays env-tunable (HERMES_STT_FINAL_BEAM_SIZE) for a future pass;
serve-before-warmup is a recommended follow-up so cold start never
blocks readiness.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BvMSXH8VH2tMWXanb8SJdf

services/hermes/voice-deployment.yaml

@@ -19,7 +19,7 @@ spec:
         app: hermes-stt
       annotations:
         ai.bstein.dev/role: private-chat-speech-to-text
-        ai.bstein.dev/model: whisper-large-v3-turbo-multilingual (tiny rolling previews)
+        ai.bstein.dev/model: whisper-large-v3-turbo greedy + proper-noun prompt (tiny rolling previews)
         ai.bstein.dev/gpu: titan-21 dedicated speech time-slice
     spec:
       runtimeClassName: nvidia
@@ -40,6 +40,13 @@ spec:
             - {name: XDG_CACHE_HOME, value: /tmp/cache}
             - {name: HERMES_STT_PORT, value: "9000"}
             - {name: HERMES_STT_MODEL, value: large-v3-turbo}
+            # Fast-and-accurate production decode: the accurate turbo model and
+            # the proper-noun priming prompt, but greedy single-temperature so
+            # startup warmup and per-utterance latency stay low (beam search
+            # made cold start ~8min on this Jetson). Beam is env-tunable if a
+            # future accuracy pass wants it.
+            - {name: HERMES_STT_FINAL_BEAM_SIZE, value: "1"}
+            - {name: HERMES_STT_FINAL_TEMPERATURE, value: "0.0"}
             - {name: HERMES_STT_CACHE, value: /opt/models/whisper}
             - {name: NVIDIA_VISIBLE_DEVICES, value: all}
             - {name: NVIDIA_DRIVER_CAPABILITIES, value: "compute,utility"}