Close write_pipe after forking

[dansimko]

The write_pipe never goes out of scope and so the last process's write_pipe is never closed by the watcher process, which results in a hang on read() from this pipe in log_watcher when the worker process is killed, and blocking all worker processes when their respective write pipe buffers are exhausted.

[thrix]

Looks like a common issue with Python multiprocessing:

Before Fix:
Parent: [read_pipe, write_pipe]   ❌ <- still holds write_pipe
Child:              [write_pipe]
EOF never happens!

After Fix:
Parent: [read_pipe] ✅
Child:              [write_pipe]
→ EOF is detected correctly when child closes or exits.

Interestingly, we do not hit it always, but this was hunting our production for quite some time now :) nice find @dansimko

[synweap15]

@dansimko great find. Thanks!