Built-in download engine stalls on mirror/CDN sites - needs read timeout and range retry

System Info

• GPT4All: 3.10.0 (desktop)
• OS: Windows 11 (Build 26100)
• Network: hf-mirror.com (China mainland HuggingFace mirror)

Reproduction

1. Open GPT4All → Discover Models
2. Download any model (e.g., Qwen2-1.5B-Instruct)
3. Observe: download starts but stalls indefinitely after a few MB

Root Cause Analysis

The built-in download engine lacks a read timeout on the HTTP connection. When downloading from mirror/CDN sites (like hf-mirror.com), the server may stop sending data mid-stream without closing the TCP connection. The download engine waits forever with no progress.

Workaround: Clicking "Cancel" then "Resume" unblocks it by sending an HTTP Range request, creating a fresh connection that works again — until the next stall.

Expected Behavior

1. Implement socket read timeout (e.g., 30s) — if no data received, close and retry with Range.
2. Use HTTP Range retry automatically on timeout, instead of requiring manual Cancel/Resume.
3. Ideally: multi-threaded segmented download + automatic retry.

Notes

Qt's QNetworkAccessManager supports read timeout via setTransferTimeout(), so this should be a straightforward fix.