[AdvaithSai]

Summary

On Windows, a connection to a closed port is retried for about 2s per address instead of failing at once, and localhost resolves to both ::1 and 127.0.0.1. So with no ComfyUI server running (the default for most users), each ComfyUIClient.is_available() probe takes about 4s. The registry probes once for each of comfyui_image, comfyui_video and comfyui_music, plus once each for image_selector and video_selector, so preflight spends about 20s waiting for a server that isn't there.

A server on the same machine completes the TCP handshake in well under a millisecond, so the health probe now uses a 0.5s connect timeout for loopback hosts (localhost, 127.0.0.0/8, ::1). Remote ComfyUI servers keep the 5s connect timeout, and the read timeout stays at 5s everywhere, so a busy local server that is slow to answer /system_stats is still detected.

Related issue

Refs #756 (one part of the "too slow" complaint; the docs and permission parts of that issue are not addressed here)

Changes

• tools/_comfyui/client.py: new _probe_timeout() that returns (0.5, 5) for loopback hosts and (5, 5) otherwise; is_available() uses it.
• tests/contracts/test_comfyui_tools.py: parametrized test covering localhost, 127.0.0.1, [::1], a LAN IP and a remote hostname.

Testing

Windows 11, Python 3.12, no ComfyUI running:

	Before	After
ComfyUIClient("http://localhost:8188").is_available()	4.08s	1.02s
each of the 5 ComfyUI-related get_status() calls	~4s	~1s
registry.provider_menu() (preflight)	41.8s	18.9s

• A stand-in HTTP server on 127.0.0.1:8188 returning 200 is still detected (0.52s via localhost, 0.02s via 127.0.0.1).
• pytest tests/contracts/test_comfyui_tools.py tests/tools/test_new_video_model_support.py tests/tools/test_provider_refresh.py → 184 passed.
• Not run on macOS/Linux. There a refused connection fails immediately, so behaviour there should be unchanged.

Checklist

• The change is focused on a single logical concern.
• I ran the relevant tests locally (make test-contracts / make test) where applicable.
• I updated docs/README if behavior or usage changed. (no usage change)
• No unrelated files (build artifacts, local config) are included in the diff.