fix(s07): resolve Windows subprocess UnicodeDecodeError with UTF-8/GBK fallback

[xiarimr]

Bug Description

On Windows, when executing commands via run_bash (such as inspecting or loading skill.md), a UnicodeDecodeError occurs in subprocess's background reader thread (_readerthread).

Trigger Scenario:
skill.md contains UTF-8 encoded text/characters. When run_bash executes CLI commands that output its content, Windows defaults to GBK (CP936) for stream decoding when text=True is enabled.

Chain of Failure:

1. UTF-8 byte sequences from skill.md fail to decode under Windows default GBK codec in _readerthread.
2. The reader thread crashes silently before returning output.
3. r.stdout evaluates to None, which triggers a secondary TypeError: unsupported operand type(s) for +: 'NoneType' and 'str' in run_bash.

Solution

1. Remove text=True from subprocess.run to keep output as raw bytes.
2. Implement a safe_decode helper function that attempts UTF-8 decoding first, with a clean fallback to GBK.

Tested Environments

• OS: Windows 10/11 (ANSI CP936 / GBK default)
• Module: s07_skill_loading