Retry upstream connection-refused with exponential backoff

A refused dial (service down/restarting) is retried with a wait that
doubles from BACKOFF_INITIAL (1s) up to BACKOFF_MAX (60s) until the
upstream answers or the client disconnects. Handles the Windows WSA
errno (10061) as well as POSIX ECONNREFUSED. compose.yaml.example now
uses host.docker.internal like the working local deployment.

cmd/gpu-turnstile/main.go

@@ -40,6 +40,8 @@ type config struct {
 	freeTimeout         time.Duration
 	warmTimeout         time.Duration
 	shutdownTimeout     time.Duration
+	backoffInitial      time.Duration
+	backoffMax          time.Duration
 	promptCaptureLimit  int64
 
 	warmModel string
@@ -76,6 +78,8 @@ func loadConfig(getenv func(string) string) (config, error) {
 		freeTimeout:         30 * time.Second,
 		warmTimeout:         2 * time.Minute,
 		shutdownTimeout:     10 * time.Second,
+		backoffInitial:      time.Second,
+		backoffMax:          time.Minute,
 		promptCaptureLimit:  64 * 1024,
 
 		logLevel: slog.LevelInfo,
@@ -107,6 +111,8 @@ func loadConfig(getenv func(string) string) (config, error) {
 		{"FREE_TIMEOUT", &cfg.freeTimeout},
 		{"WARM_TIMEOUT", &cfg.warmTimeout},
 		{"SHUTDOWN_TIMEOUT", &cfg.shutdownTimeout},
+		{"BACKOFF_INITIAL", &cfg.backoffInitial},
+		{"BACKOFF_MAX", &cfg.backoffMax},
 	} {
 		if err := envDuration(getenv, e.name, e.dst); err != nil {
 			return cfg, err
@@ -185,6 +191,8 @@ func main() {
 		HistoryPollInterval: cfg.historyPollInterval,
 		FreeTimeout:         cfg.freeTimeout,
 		WarmTimeout:         cfg.warmTimeout,
+		BackoffInitial:      cfg.backoffInitial,
+		BackoffMax:          cfg.backoffMax,
 		PromptCaptureLimit:  cfg.promptCaptureLimit,
 		WarmModel:           cfg.warmModel,
 	})