fix(dictation): typed transcription session, sync editor before merge, block sends during upload

- Use OpenAI typed transcription session format (type: "transcription")
  instead of legacy realtime fields that would fail or produce no transcripts
- Sync editor content via syncContentRef before merging dictation text so
  manually typed prefixes are preserved when dictation starts
- Read send-blocked state from refs at transcript time so uploads prevent
  auto-submit from clearing the composer

Signed-off-by: klopez4212 <klopez4212@gmail.com>

crates/buzz-relay/src/api/transcribe.rs

@@ -82,8 +82,7 @@ pub async fn create_transcribe_session(
         .header("Content-Type", "application/json")
         .json(&serde_json::json!({
             "session": {
-                "model": "gpt-4o-mini-realtime-preview",
-                "modalities": ["text"],
+                "type": "transcription",
                 "input_audio_transcription": {
                     "model": model,
                 },