fix(websocket): fallback to HTTP/1.1 when H2 CONNECT is unavailable

[mcollina]

Summary

Fix websocket handshakes when TLS negotiates HTTP/2 but the server does not advertise RFC 8441 extended CONNECT support.

Instead of failing immediately, retry the opening handshake on a fresh HTTP/1.1-only connection. This preserves websocket support for allowHTTP1 HTTP/2 servers and other deployments that only support websocket upgrade over HTTP/1.1.

Changes

• add a regression test for websocket fallback to HTTP/1.1 upgrade
• retry websocket handshakes over HTTP/1.1 when H2 extended CONNECT is unavailable
• plumb the HTTP/1.1-only retry through Agent and proxy TLS connectors

Testing

• npx eslint lib/dispatcher/client.js lib/dispatcher/proxy-agent.js lib/dispatcher/agent.js lib/web/fetch/index.js test/websocket/opening-handshake.js
• node --test test/websocket/opening-handshake.js

[codecov-commenter]

Codecov Report

❌ Patch coverage is 96.58120% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 92.96%. Comparing base (490cbc6) to head (92c7b8f).

Files with missing lines	Patch %	Lines
lib/web/fetch/index.js	95.95%	8 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4966   +/-   ##
=======================================
  Coverage   92.95%   92.96%           
=======================================
  Files         111      111           
  Lines       35821    35864   +43     
=======================================
+ Hits        33299    33342   +43     
  Misses       2522     2522           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

@pimterry PTAL

[pimterry]

I'm not super familiar with the Undici internals, but they look roughly sensible to me - the test definitely looks like the right end behaviour 👍