Fix infinite WebSocket loops in test_server causing 6 hour CI timeouts

[endolith]

Describe the changes you have made:

Fix infinite WebSocket loops in test_server causing CI timeouts

Replace 6 infinite while True loops with retry-limited helper function.

PROBLEM:

• GitHub Actions tests were timing out after 6 hours on PRs (https://github.com/openinterpreter/open-interpreter/actions/runs/18751966322 https://github.com/openinterpreter/open-interpreter/actions/runs/18752011550 https://github.com/openinterpreter/open-interpreter/actions/runs/18751990723/ https://github.com/openinterpreter/open-interpreter/actions/runs/18751982247/ etc)
• PRs don't have access to GitHub secrets(?), causing authentication failures
• test_server() contained 6 infinite while True loops which waited for 'complete' status messages that never arrived
• Server never sends 'complete' status on auth failure, causing indefinite hangs
• No retry limits or timeouts caused indefinite blocking

SOLUTION:

• Add wait_for_websocket_complete() helper function with max 5 attempts
• Replace all 6 infinite loops with calls to helper function
• Fail fast with descriptive error messages instead of hanging

This prevents 6-hour CI timeouts and provides clear debugging information when WebSocket communication fails.

Reference any relevant issues (e.g. "Fixes #000"):

Pre-Submission Checklist (optional but appreciated):

• [ ] I have included relevant documentation updates (stored in /docs)
• [x] I have read docs/CONTRIBUTING.md
• [x] I have read docs/ROADMAP.md

OS Tests (optional but appreciated):

• [ ] Tested on Windows
• [ ] Tested on MacOS
• [ ] Tested on Linux

[endolith]

(Ironically this PR triggers the same problem: https://github.com/openinterpreter/open-interpreter/actions/runs/18819992234 😅)

[Notnaton]

Should we nuke the tests? It needs an API key for the tests anyways so nothing will run.

[endolith]

They are running once it's merged to main: https://github.com/openinterpreter/open-interpreter/actions/runs/19034913261

So Killian must have put API keys into GitHub secrets, but feature branches don't have access to the secrets for security reasons, if I understand correctly. (So people can't submit PRs that steal the secrets. AI told me it's possible to set it up for manual triggering, too, like you could press a button to test with secrets after confirming the PR code is safe to access them.)

I also had some changes that skip API tests for feature branches, but I'm not sure of best practice there. But with this PR those should fail quickly at least.