Fix subprocess stream decoding for split UTF-8 output

[0xTaoZ]

Fixes #574.

This updates the subprocess wrapper to drain stdout and stderr concurrently while decoding progress stdout with an incremental UTF-8 decoder. That prevents split multibyte sequences from being surfaced as replacement characters while preserving the existing progress update behavior.

Tests added:

• split multibyte stdout with log_stdout=True
• split multibyte stderr on subprocess failure

Local checks:

• .venv/bin/python -m pytest test/test_subprocess.py -q
• .venv/bin/python -m ruff check pip_audit/_subprocess.py test/test_subprocess.py
• .venv/bin/python -m mypy pip_audit/_subprocess.py
• .venv/bin/python -m ruff format --check pip_audit/_subprocess.py test/test_subprocess.py
• git diff --check