fix: clear the CI build annotations

[devopvoid]

Fixes the annotations CI attaches to every build (e.g. run 36268227980) and the libc++ warning flooding the Linux build logs. One commit per issue.

Node.js 20 deprecation (ci:)

GitHub forces actions that target Node.js 20 onto Node.js 24 and flags every job ("The following actions target Node.js 20 but are being forced to run on Node.js 24: actions/cache@v4, actions/upload-artifact@v4"). Every action still on Node.js 20 moves to its current Node.js 24 major:

Action	From	To
actions/cache	v4	v6
actions/upload-artifact	v4	v7
actions/download-artifact	v4	v8
actions/setup-python	v5	v7
actions/configure-pages	v5	v6
actions/deploy-pages	v4	v5
actions/upload-pages-artifact	v4	v5 (wraps upload-artifact v7)

None of their breaking changes affect how the workflows use them: artifacts are downloaded by name/pattern (download-artifact v5 only changed download-by-ID paths), download-artifact v8 fails on hash mismatches instead of warning, and pip-install (removed in setup-python v7) is not used. checkout, setup-java, setup-node and git-cliff-action already run on Node.js 24.

[module] module not found: webrtc.java.media (build:)

webrtc.java exports dev.onvoid.webrtc.internal only to webrtc.java.media. That module depends on webrtc.java, so it is never observable while webrtc.java compiles, and javac warns on every build. The export is intentional and still takes effect at run time, so the lint is suppressed on the module declaration (@SuppressWarnings("module")) instead of for the whole build.

logToDebug at ERROR (fix:)

This came from an earlier test, not from LoggingTests. RTCDataChannelSendObserver reported an exception thrown by the observer with ExceptionDescribe, which prints Exception in thread "Thread-N" straight to native stderr without a newline and the stack trace to System.err. RTCDataChannelSendCompletionTests throws from an observer on purpose, so that prefix sat on an unterminated line until LoggingTests' native debug line landed on it, and setup-java's java problem matcher (^Exception in thread "(.*)" (.*): (.*)$) reported the result as an error.

The observer now hands the exception to the calling thread's uncaught exception handler, as the JVM does for an exception escaping a Java thread, so the report goes through System.err in one piece and applications can install their own handler. The two tests that throw on purpose capture the exception in a default handler and assert it arrived; against the previous native library they time out.

"Libc++ only supports Clang 21 and later" (ci:)

prepare-linux installed a Clang 20 development snapshot (llvmorg-20-init-9764) into /opt/clang, which the Linux toolchain files compile the natives with. The JNI code compiles against the libc++ headers installed by the WebRTC build, and those of branch-heads/7977 only support Clang 21 and later, so every Linux build printed the warning once per translation unit (294 times on linux_arm alone).

prepare-linux now installs the package that tools/clang/scripts/update.py pins on that branch (llvmorg-23-init-19482-g53d18800-1), the compiler WebRTC itself is built with. One Linux_x64 package serves all three Linux jobs, since the arm jobs cross-compile on x64 runners. A comment in the action notes that the pin has to follow webrtc.branch. No build uses -Werror, so any new warnings from the newer Clang would show up in the log rather than fail the build. This change can only be verified by CI; there is no Linux build here.

Testing

Tested locally on Windows x86_64: mvn -pl webrtc test passes (175 tests) with no output line matching any of setup-java's problem matchers, and RTCDataChannelSendCompletionTests passes under -Pjni-check with no FATAL ERROR in native method.