fix: wire DockerSocket config and keep VM alive after entrypoint exits

Two fixes for macOS development:

1. Wire cfg.Build.DockerSocket into builds.Config so the config file
   value (e.g. Colima socket path) is actually used instead of always
   falling back to /var/run/docker.sock.

2. Restore pre-PR#99 behavior of keeping the VM alive after the
   entrypoint exits by waiting on the guest-agent. PR#99 changed init
   to immediately power off the VM when the entrypoint exits, which
   breaks images like alpine:latest whose CMD is /bin/sh — the shell
   gets no stdin and exits instantly, killing the VM before anyone can
   `hm exec` into it. The guest-agent keeps the VM alive and accessible
   until an explicit stop/delete.

Author: rgarcia

lib/providers/providers.go

@@ -289,6 +289,7 @@ func ProvideBuildManager(p *paths.Paths, cfg *config.Config, instanceManager ins
 		RegistryCACert:      registryCACert,
 		DefaultTimeout:      cfg.Build.Timeout,
 		RegistrySecret:      cfg.JwtSecret, // Use same secret for registry tokens
+		DockerSocket:        cfg.Build.DockerSocket,
 	}
 
 	// Configure secret provider (use NoOpSecretProvider as fallback to avoid nil panics)

lib/system/init/mode_exec.go

@@ -137,6 +137,16 @@ func runExecMode(log *Logger, cfg *vmconfig.Config) {
 	// discovers the VM has stopped (socket gone -> Stopped state).
 	log.Info("hypeman-init:entrypoint", formatExitSentinel(exitCode, exitMsg))
 
+	// Keep VM alive by waiting on the guest-agent. The agent runs forever,
+	// so the VM stays accessible via `hm exec` even after the entrypoint
+	// exits (e.g. alpine's /bin/sh exits immediately without stdin).
+	// The VM is shut down when a stop/delete signal arrives (forwarded above)
+	// or when the guest-agent's Shutdown RPC sends SIGTERM to PID 1.
+	if agentCmd != nil && agentCmd.Process != nil {
+		log.Info("hypeman-init:entrypoint", "entrypoint finished, waiting on guest-agent to keep VM alive")
+		agentCmd.Wait()
+	}
+
 	// Clean shutdown: use reboot(POWER_OFF) instead of syscall.Exit to avoid
 	// kernel panic ("Attempted to kill init!"). This cleanly terminates the VM
 	// and causes the hypervisor process to exit on the host.