feat: add Codec2 voice messages over LoRa mesh (iOS/macOS only)

Push-to-talk voice messaging using the Codec2 ultra-low-bitrate speech
codec, transmitted as V: prefixed packets over the existing MeshCore
LoRa mesh pipeline.

Voice recording UI (long-press send button or + menu) is gated behind
Platform.isIOS || Platform.isMacOS since the `record` package only
supports microphone capture on those platforms in this build.

Key changes:
- VoiceRecorderService: streams 8kHz mono PCM chunks via `record` package
- VoicePlayerService: decodes Codec2 bytes to WAV and plays via audioplayers
- VoiceCodecService: async Codec2 encode/decode in background isolates
- VoiceProvider: reassembles multi-packet sessions, drives playback
- VoiceMessageBubble: shows packet progress, play/stop controls
- MessagesProvider: detects V: prefix, routes to VoiceProvider
- MessagesTab: PTT long-press gesture + recording indicator (iOS/macOS)
- Message model: isVoice + voiceId fields for session tracking
- Auto-selects codec mode from radio bandwidth (700C/1200/1300 bps)
- codec2_flutter + meshcore_client switched from path to git deps

lib/models/message.dart

@@ -9,6 +9,20 @@ export 'package:meshcore_client/meshcore_client.dart'
 import 'package:flutter/foundation.dart';
 import 'package:meshcore_client/meshcore_client.dart';
 import 'sar_marker.dart';
+import '../utils/voice_message_parser.dart';
+
+extension MessageVoiceExtension on Message {
+  /// True when this message is a voice recording (`V:` text or binary voice).
+  bool get isVoiceMessage => isVoice;
+
+  /// Parses the voice packet mode from the stored [voiceId] session info.
+  /// Returns null for non-voice messages.
+  VoicePacketMode? get voicePacketMode {
+    if (!isVoice || text.isEmpty) return null;
+    final pkt = VoicePacket.tryParseText(text);
+    return pkt?.mode;
+  }
+}
 
 extension MessageSarExtension on Message {
   /// Infer the [SarMarkerType] from stored SAR fields.