fix(windows): decode Windows bash output encodings

[yuczzzzz]

Why

On Windows, some bash commands produce output bytes in a non-UTF-8 encoding. Decoding those bytes as UTF-8 produces garbled text, making command output difficult to read and use.

Goal

Ensure bash output on Windows is correctly decoded as Unicode text when it uses a supported non-UTF-8 encoding, while preserving existing UTF-8 behavior and non-Windows behavior.

Summary

• Preserve raw byte chunks from bash stdout and stderr, then decode each stream independently after the process exits.
• On Windows, validate UTF-8 strictly before attempting encoding detection. For invalid UTF-8, accept only chardet's first candidate when confidence is at least 80 and iconv-lite supports it.
• Fall back to tolerant UTF-8 decoding when detection is absent, low-confidence, unsupported, or throws.
• Keep the existing combined raw-byte output limit, timeout handling, redaction, UTF-8 truncation, and external bash tool interface.
• Add direct runtime dependencies for chardet and iconv-lite.
• Add encoding smoke coverage for UTF-8, GBK, GB18030, fallback paths, split multi-byte output, truncated UTF-8, and Windows mixed stdout/stderr encoding.

Security Impact

This change affects shell-output decoding only. It does not change bash command permissions, execution modes, MCP parameters, environment variables, or result fields. Output retention remains bounded by the existing raw-byte limit. Low-confidence or unsupported encoding detection falls back to the existing UTF-8 behavior.

Validation

• npm run build
• npm run smoke
• npm run stress
• Update README.md or CHANGELOG.md for this behavior change.