Revert "Add WebKit to e2e test matrix (#37298)"

[silverwind]

Reverts: #37298
Ref: https://github.com/go-gitea/gitea/actions/runs/24661464168/job/72108324223?pr=37312

WebKit on Linux has a long history of flakiness in Playwright CI runs, and the exact "WebKit encountered an internal error" class of failures has been reported repeatedly and closed without a real fix (typically stale/no-repro, or worked around by retries):

• [Bug]: Error: page.goto: WebKit encountered an internal error occurs randomly on CI microsoft/playwright#34450
• [Bug]: Failed to preconnect to https://use.typekit.net/eqc5doy.css. Error: WebKit encountered an internal error microsoft/playwright#35773
• [internal]: webkit linux is flaky with ignoreHTTPSErrors microsoft/playwright#35870
• [BUG] HSTS with Redirection is broken in WebKit on Linux microsoft/playwright#35293
• [Bug]: Navigation API on webkit linux silently doesn't work microsoft/playwright#38838

Keeping chromium and firefox in the e2e matrix.

---

This PR was written with the help of Claude Opus 4.7

[silverwind]

If we have MacOS runner resources, it may be an option as WebKit seems much more stable on MacOS than Linux, otherwise I opt for the revert.

[silverwind]

For future reference, I made Claude reproduce this failure inside docker, here is its verdict:

---

Reproduced on iteration 4/30. Root cause is visible in browser stderr right before the playwright error:

pw:browser [pid=43942][err] malloc_consolidate(): unaligned fastbin chunk detected
pw:browser [pid=43942][err] ERROR: WebKit encountered an internal error. This is a WebKit bug.
pw:browser [pid=43942][err] /home/pwuser/webkit/Source/WebKit/WebProcess/Network/WebLoaderStrategy.cpp(668)
                            : void WebKit::WebLoaderStrategy::internallyFailedLoadTimerFired()

What happened: The WebContent (WebProcess) died from glibc heap corruption (malloc_consolidate: unaligned fastbin chunk detected). Playwright saw the kill, hit WebLoaderStrategy::internallyFailedLoadTimerFired on the browser side, and surfaced that as "WebKit encountered an internal error" on the next page.goto. The specific test is incidental — CI hit it on login.test.ts, this run caught it on issue-project.test.ts.

This confirms the class of bug cited in the revert commit: a genuine WebKit C++ heap bug, not a gitea-side problem. The revert was the right call.

Harness

• Worktree: /tmp/gitea-webkit-repro (branch webkit-repro, cherry-pick of 6f761219b2)
• Dockerfile + run-repro.sh: /tmp/webkit-repro-harness/
• Image: gitea-webkit-repro (Ubuntu 24.04, Go 1.26.2, Node 24, pnpm 10.33.0)
• Output: /tmp/webkit-repro-out/ — run-004.log has the full crash, trace artefacts in run-004.e2e-output/
• Repro rate: ~1 in 4 iterations here (arm64 + orbstack). Rerun with docker run --rm --shm-size=2g --cpus=4 --ulimit core=-1 -v /tmp/gitea-webkit-repro:/src -v /tmp/webkit-repro-out:/out gitea-webkit-repro 30.