fix(console): resolve StderrCapture memory leak and fd-2 restoration - #435

Summary

Fixes #433 — memory leak while app is idle caused by StderrCapture producing runaway _Block_copy allocations.

Four bugs in StderrCapture are fixed:

1. readabilityHandler leak (#433, primary)

The readabilityHandler dispatched captureQueue.async { … } and read fileHandle.availableData inside the async block. The read source stayed signalled because availableData wasn't consumed in the handler callback itself, so it re-fired immediately — enqueuing a new dispatch block per callback and leaking unbounded _Block_copy allocations (48/64-byte blocks matching the Instruments trace).

Fix: Read availableData synchronously inside the readabilityHandler to clear the dispatch source signal. Only the heavier string parsing/forwarding is dispatched to processingQueue.

2. originalDescriptor aliasing (infinite recursion)

originalDescriptor stored FileHandle.standardError.fileDescriptor (fd 2) directly. After dup2 redirected fd 2 onto the capture pipe, writeDirectlyToOriginalStderr wrote back into the capture loop — infinite recursion.

Fix: dup() the real stderr fd before redirect so originalDescriptor is an owned copy that survives the redirect. Close it in stopCapturingInternal and on error paths.

3. stopCapturingInternal freopen breaks fd 2 (found during testing)

freopen("/dev/stderr", "a", stderr) closes fd 2 first, then tries to open /dev/fd/2 — which is now closed, so it fails with EBADF and leaves fd 2 permanently invalid. This broke the next startCapturing()'s dup(2) (returned -1) and silently broke all post-stop stderr output (NSLog, crash logs, OS-level writes).

Fix: Replace freopen with dup2(originalDescriptor, fd2) + clearerr(stderr) + setvbuf(stderr, nil, _IOLBF, 0). The dup2 restores fd 2 to real stderr; the C stderr FILE* stream still references fd 2, so it writes to the right destination. clearerr resets error state; setvbuf restores line-buffered mode.

4. stopCapturingInternal fd-2 restoration ordering

The original code had no dup2 restore before freopen — fd 2 stayed pointing at the capture pipe after stop, so stderr stayed redirected.

Fix: dup2(originalDescriptor, FileHandle.standardError.fileDescriptor) to restore real stderr (now combined with fix #3 — no freopen at all).

Test plan

All three regression tests pass (previously 2/3 skipped due to the freopen bug found and fixed in this PR):

• StderrCaptureTests.testSingleStderrWriteProducesOneConsoleEntry — Passed (2.4s). Verifies one stderr write produces ≤2 console entries, not thousands (the leak symptom).
• StderrCaptureTests.testStderrPassthroughDoesNotRecurse — Passed (2.3s). Writes a [marker] uuid line; the ] triggers writeDirectlyToOriginalStderr. Asserts the bare uuid fragment does NOT appear as its own console entry (the recursion-fix path).
• StderrCaptureTests.testStopRestoresStderr — Passed (2.3s). After stop, writes to stderr and asserts they do NOT appear in the console (fd 2 restored).
• xcodebuild test on Example scheme — ** TEST SUCCEEDED **

Tests use setUp to stop app-launched capture so every test starts from a known-stopped state and exercises a real startCapturing → startCapturingInternal path through the fixed code.

Closes #433