feat(voice): self-hosted speech recognition via Whisper (STT_*)

- STT_URL switches voice input to an OpenAI-compatible Whisper server
  (e.g. speaches); without it the browser's recognition stays. Configured
  by environment only, with limits for upload size, parallel jobs, hourly
  budget per user and timeout. The demo always uses the browser.
- POST /voice/transcribe forwards recordings and returns the text; nothing
  is stored. On startup MindSift waits for the server and asks it to
  download the model if missing.
- Admin card shows the active mode and server state, with a test recording.
- docker-compose: optional "whisper" service (speaches 0.8.3 CPU, profile
  "whisper", models in ./whisper).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

internal/api/bodylimit.go

@@ -12,8 +12,10 @@ import (
 // with huge uploads. Note sync and imports need more; their handlers apply the same limits again.
 const defaultMaxBody = 2 << 20
 
-func bodyLimitFor(path string) int64 {
+func (s *Server) bodyLimitFor(path string) int64 {
 	switch {
+	case path == "/api/v1/voice/transcribe":
+		return int64(s.cfg.Speech.MaxUploadMB)<<20 + 1<<20
 	case strings.HasPrefix(path, "/api/v1/import/"):
 		return storage.MaxImportUploadBytes + 1<<20
 	case strings.HasPrefix(path, "/api/v1/docs/") && strings.HasSuffix(path, "/sync"):
@@ -22,10 +24,10 @@ func bodyLimitFor(path string) int64 {
 	return defaultMaxBody
 }
 
-func bodyLimit(next http.Handler) http.Handler {
+func (s *Server) bodyLimit(next http.Handler) http.Handler {
 	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		if r.Body != nil && r.Body != http.NoBody {
-			r.Body = http.MaxBytesReader(w, r.Body, bodyLimitFor(r.URL.Path))
+			r.Body = http.MaxBytesReader(w, r.Body, s.bodyLimitFor(r.URL.Path))
 		}
 		next.ServeHTTP(w, r)
 	})