feat(voice): add audio-model transcription support

- Add `AudioModelTranscriber` for model-based audio transcription via LLM providers
- Support selecting a transcription model with `voice.model_name` in config
- Keep Groq transcription as a fallback and move it into dedicated files with focused tests
- Serialize `data:audio/...` media as input_audio for OpenAI-compatible providers
- Improve transcription logging by rendering error fields as strings
- Add coverage for transcriber detection, audio-model behavior, provider audio serialization, and Groq transcription

Fixes #1890.

pkg/logger/logger.go

@@ -256,6 +256,8 @@ func appendFields(event *zerolog.Event, fields map[string]any) {
 	for k, v := range fields {
 		// Type switch to avoid double JSON serialization of strings
 		switch val := v.(type) {
+		case error:
+			event.Str(k, val.Error())
 		case string:
 			event.Str(k, val)
 		case int:

pkg/logger/logger_test.go

@@ -1,7 +1,12 @@
 package logger
 
 import (
+	"bytes"
+	"encoding/json"
+	"errors"
 	"testing"
+
+	"github.com/rs/zerolog"
 )
 
 func TestLogLevelFiltering(t *testing.T) {
@@ -337,3 +342,26 @@ func TestSetLevelFromString(t *testing.T) {
 		t.Errorf("after SetLevelFromString(\"FATAL\"): GetLevel() = %v, want FATAL", got)
 	}
 }
+
+func TestAppendFields_ErrorUsesErrorString(t *testing.T) {
+	var buf bytes.Buffer
+	l := zerolog.New(&buf)
+
+	event := l.Info()
+	appendFields(event, map[string]any{"error": errors.New("transcription request failed")})
+	event.Msg("test")
+
+	lines := bytes.Split(bytes.TrimSpace(buf.Bytes()), []byte("\n"))
+	if len(lines) == 0 {
+		t.Fatal("expected log output, got none")
+	}
+
+	var got map[string]any
+	if err := json.Unmarshal(lines[0], &got); err != nil {
+		t.Fatalf("unmarshal log line: %v", err)
+	}
+
+	if got["error"] != "transcription request failed" {
+		t.Fatalf("error field = %#v, want %q", got["error"], "transcription request failed")
+	}
+}