Extend retry backoff; rework logging (LOGLEVEL, request lines, colors)

Backoff now covers all dial-phase errors (refused, timeout, DNS), TLS
handshake failures, and 5xx responses with replayable bodies; streamed
POSTs are never replayed to avoid duplicate work. First retry of an
episode logs at WARN, subsequent attempts at INFO.

LOGLEVEL (LOG_LEVEL kept as alias) now defaults to warn: startup logs
version plus every setting; INFO adds one line per incoming request and
per response with status/duration, ANSI-colored in text mode (bypasses
slog's escaping so colors render in docker compose logs); NO_COLOR or
LOG_FORMAT=json disables colors.

cmd/gpu-turnstile/main.go

@@ -82,7 +82,7 @@ func loadConfig(getenv func(string) string) (config, error) {
 		backoffMax:          time.Minute,
 		promptCaptureLimit:  64 * 1024,
 
-		logLevel: slog.LevelInfo,
+		logLevel: slog.LevelWarn,
 	}
 	for _, e := range []struct {
 		name string
@@ -125,10 +125,15 @@ func loadConfig(getenv func(string) string) (config, error) {
 		}
 		cfg.promptCaptureLimit = n
 	}
-	if v := getenv("LOG_LEVEL"); v != "" {
+	// LOGLEVEL is the canonical spelling; LOG_LEVEL is kept as an alias.
+	logLevelValue := getenv("LOGLEVEL")
+	if logLevelValue == "" {
+		logLevelValue = getenv("LOG_LEVEL")
+	}
+	if logLevelValue != "" {
 		var level slog.Level
-		if err := level.UnmarshalText([]byte(v)); err != nil {
-			return cfg, fmt.Errorf("LOG_LEVEL: %w", err)
+		if err := level.UnmarshalText([]byte(logLevelValue)); err != nil {
+			return cfg, fmt.Errorf("LOGLEVEL: %w", err)
 		}
 		cfg.logLevel = level
 	}
@@ -157,12 +162,29 @@ func main() {
 	log := slog.New(handler)
 	slog.SetDefault(log)
 
-	log.Info("starting gpu-turnstile",
+	// The startup line carries the version and every setting and is emitted
+	// at WARN so it is visible even with the default (quiet) log level.
+	log.Log(context.Background(), slog.LevelWarn, "starting gpu-turnstile",
 		"version", version,
 		"listen_ollama", cfg.listenOllama,
 		"listen_comfy", cfg.listenComfy,
 		"ollama_url", cfg.ollamaURL,
 		"comfy_url", cfg.comfyURL,
+		"unload_timeout", cfg.unloadTimeout,
+		"job_timeout", cfg.jobTimeout,
+		"llm_wait_timeout", cfg.llmWaitTimeout,
+		"unload_poll_interval", cfg.unloadPollInterval,
+		"history_poll_interval", cfg.historyPollInterval,
+		"probe_timeout", cfg.probeTimeout,
+		"free_timeout", cfg.freeTimeout,
+		"warm_timeout", cfg.warmTimeout,
+		"shutdown_timeout", cfg.shutdownTimeout,
+		"backoff_initial", cfg.backoffInitial,
+		"backoff_max", cfg.backoffMax,
+		"prompt_capture_limit", cfg.promptCaptureLimit,
+		"warm_model", cfg.warmModel,
+		"log_level", cfg.logLevel,
+		"log_format", map[bool]string{true: "json", false: "text"}[cfg.logJSON],
 	)
 
 	ollamaClient, err := ollama.New(cfg.ollamaURL, log)
@@ -184,6 +206,7 @@ func main() {
 		Comfy:               comfyClient,
 		Metrics:             metrics.New(),
 		Log:                 log,
+		LogColor:            !cfg.logJSON && os.Getenv("NO_COLOR") == "",
 		LLMWaitTimeout:      cfg.llmWaitTimeout,
 		UnloadTimeout:       cfg.unloadTimeout,
 		JobTimeout:          cfg.jobTimeout,