fix: stop logger teardown from closing process stdout

[raulmol]

Fixes #30.

setup_logger wraps sys.stdout.buffer in a new TextIOWrapper per call; after clean_logger drops the RichHandler, garbage collection of that wrapper closes the underlying buffer — the process's real stdout. The next setup_logger call in the same process (organize stage after setup stage) then raises ValueError: I/O operation on closed file, and run_organize's own console.print crashes with the same error while reporting it, masking the root cause.

Change

launch/utilities/logger.py only: create the UTF-8-wrapped rich Console once at module level (_get_shared_console) and reuse it for every RichHandler, so the wrapper is never collected and stdout stays open across logger setup/teardown cycles. Behavior (UTF-8 re-encoding with errors="replace", soft wrap) is unchanged.

Verification

import gc, sys, tempfile
from pathlib import Path
from launch.utilities.logger import setup_logger, clean_logger

d = Path(tempfile.mkdtemp())
for i in range(2):
    lg = setup_logger(f"inst{i}", [d / f"l{i}.log"], printing=True)
    lg.info("cycle %d", i)
    clean_logger(lg)
    gc.collect()

sys.stdout.buffer.write(b"stdout alive\n")  # raised ValueError before this fix

Also verified end-to-end on a private monorepo task: before the fix, the organize stage failed immediately after a successful setup stage with I/O operation on closed file; with the fix it proceeds into the organize agents.

[raulmol]

@microsoft-github-policy-service agree