Emit 'exit' immediately on EXIT byte (~38x exec speedup)

[jhgaylor]

Summary

In the non-TTY exec path, the server sends a single EXIT byte over the WebSocket and then holds the connection open ~5 seconds before initiating its own close handshake. The current handler calls this.close() on EXIT and only emits the 'exit' event from handleClose — which fires when the WS actually closes — so every exec eats the server's ~5s wait before `execFile`'s promise resolves.

This is the same bug we just fixed in the Python SDK (superfly/sprites-py#26). The Elixir SDK (sprites-ex/lib/sprites/command.ex#L317-L324) does not have it — it sends a CLOSE frame and immediately tears down the TCP via `:gun.close`, never waiting for the server's reciprocal frame.

Fix

Emit `'exit'` synchronously when the EXIT byte arrives, then issue the WS close as fire-and-forget. `handleClose` is already guarded by `if (!this.done)` so it won't double-emit.

 case StreamID.Exit:
   this.exitCode = payload.length > 0 ? payload[0] : 0;
+  if (!this.done) {
+    this.done = true;
+    this.emit('exit', this.exitCode);
+  }
   this.close();
   break;

Impact

Side-by-side bench against a warm sprite running `cat /tmp/x` (n=2, p50):

	exec_first	exec_warm_avg
before	~6,257ms	~5,463ms
after	75ms	127ms

That's a ~80× speedup on cold and ~43× on warm. The remaining ~75ms is mostly TLS handshake + WS upgrade + actual command execution, which puts `sprites-js` on parity with the other SDKs:

SDK	exec_first p50	exec_warm_avg p50
sprites-js (this PR)	75ms	127ms
sprites-py (with #26)	106ms	128ms
sprites-go	183ms	190ms
sprites-ex	177ms	198ms

Behavior change

The `'exit'` event is now emitted before the WS finishes closing. Consumers that explicitly waited for the WS `close` event will see no change — `handleClose` is unchanged and still runs. Anything reading `exitCode`/the `exit` event keeps working but resolves ~5s sooner.

The CLOSE frame is still sent — the server sees a clean close — we just don't block the JS event loop waiting for the server's reciprocal CLOSE.