[BUG] UnicodeDecodeError on Windows non-UTF-8 consoles (CP950/CP936) when running apm init and other commands

[edenfunf]

Summary

On Windows systems configured with non-UTF-8 console encodings (CP950, CP936, CP932, etc.), several APM commands fail with an unhandled exception in a background thread:

Exception in thread Thread-N (_readerthread):
Traceback (most recent call last):
  File "threading.py", line 1043, in _bootstrap_inner
  File "subprocess.py", line 1615, in _readerthread
UnicodeDecodeError: 'cp950' codec can't decode byte 0x85 in position 4: illegal multibyte sequence

Steps to reproduce

1. Use Windows with a Traditional Chinese (CP950) or Simplified Chinese (CP936) locale
2. Run apm init <project-name>

The error appears in the terminal and the command may abort or produce incomplete output.

Root cause

subprocess.run(..., text=True) without an explicit encoding argument uses the system locale encoding to decode subprocess output. On CP950 systems, when git emits UTF-8 byte sequences (e.g. \xc2\x85), the CP950 codec raises UnicodeDecodeError in the _readerthread that reads the subprocess pipe.

The existing _configure_encoding() sets PYTHONIOENCODING=utf-8 and reconfigures sys.stdout/sys.stderr, but this does not affect how the calling Python process decodes output read back from subprocess pipes -- that path uses the locale default independently.

Affected call sites include _auto_detect_author(), _auto_detect_description(), git ls-remote validation in install.py, token_manager.py, github_downloader.py, policy/discovery.py, version.py, and all runtime adapters (codex_runtime.py, copilot_runtime.py, llm_runtime.py, manager.py).

Expected behavior

Commands complete normally. Non-ASCII output from git (e.g. a username in Chinese characters) is handled correctly.

Environment

• Windows 11 with Traditional Chinese locale (CP950)
• APM 0.8.10
• Python 3.13