feat(dictation): transcribe after recording instead of live

Parakeet is an offline model trained on whole utterances, so re-decoding
the growing buffer to animate a live transcript cost O(n^2) work for a
result the final decode replaced. Sessions now decode once per committed
segment, and the composer shows a scrolling waveform of the mic level
instead of running text.

Long dictations split at a pause once past 60s (hard cap 90s) instead of
on a blind 15s timer, so cuts no longer land mid-word. Committed segments
decode while the user is still speaking: a 185s dictation returns 4.1s
after stop instead of 11.0s, with identical text (816 vs 817 words).

Also fixes two ways the stream manager could silently drop transcribed
audio. It now counts the commits it issued instead of trusting the
session's echoed events, so a commit still in flight when the client
finishes can no longer be left out of the final text. And segment
byte/peak accounting is reset where the commit is issued rather than when
the event arrives, which could mistake the tail of a dictation for
silence and clear it.

packages/web/server/lib/dictation/local/worker-process.js

@@ -17,7 +17,7 @@
 
 import {
   SherpaOfflineRecognizerEngine,
-  SherpaRealtimeTranscriptionSession,
+  SherpaSegmentTranscriptionSession,
 } from './sherpa-recognizer.js';
 import { SherpaTtsEngine } from './sherpa-tts.js';
 import { getLocalSttModelDir, getLocalSttModelSpec } from './model-catalog.js';
@@ -126,7 +126,7 @@ async function handleRequest(message) {
     case 'session.create': {
       cleanupSession(message.sessionId);
       const engine = getEngine(message.modelsDir, message.modelId);
-      const session = new SherpaRealtimeTranscriptionSession({ engine });
+      const session = new SherpaSegmentTranscriptionSession({ engine });
       session.on('committed', (payload) => {
         sendToParent({ type: 'session.committed', sessionId: message.sessionId, payload });
       });