fix(cmd): every local engine call carries its own deadline (review F4)

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

cmd/tapir/uploads.go

@@ -108,7 +108,12 @@ func buildTranscriber(cfg config.Config) uploads.Transcriber {
 	// No overall client timeout: a 90 minute recording takes minutes. The deadline is
 	// the worker's per-job context.
 	client := &http.Client{}
-	primary := transcribe.NewGateway(cfg.GatewayURL, cfg.GatewayKey, cfg.TranscribeModel, client)
+	// Each LOCAL engine call gets its own deadline: a wedged engine does not answer, and
+	// without one it used the whole job before the fallback could run.
+	local := func(model string) transcribe.Backend {
+		return transcribe.WithTimeout(transcribe.NewGateway(cfg.GatewayURL, cfg.GatewayKey, model, client), cfg.TranscribeEngineTimeout)
+	}
+	primary := local(cfg.TranscribeModel)
 	var fallback transcribe.Backend
 	if cfg.TranscribeFallbackModel != "" && cfg.TranscribeFallbackModel != cfg.TranscribeModel {
 		fallback = transcribe.NewGateway(cfg.GatewayURL, cfg.GatewayKey, cfg.TranscribeFallbackModel, client)
@@ -117,7 +122,7 @@ func buildTranscriber(cfg config.Config) uploads.Transcriber {
 		return transcribe.NewChain(primary, fallback)
 	}
 
-	base := transcribe.NewGateway(cfg.GatewayURL, cfg.GatewayKey, cfg.ReconcileBaseModel, client)
+	base := local(cfg.ReconcileBaseModel)
 	// Temperature 0 and a short reply: the adjudicator's accuracy was measured that way.
 	chat := llm.New(cfg.GatewayURL, cfg.GatewayKey, cfg.ReconcileModel, adjudicatorTimeout,
 		llm.WithTemperature(0), llm.WithMaxTokens(200))