Edit tool corrupts multi-byte UTF-8 characters at replacement string boundary

Description

The Edit tool occasionally corrupts multi-byte UTF-8 characters (e.g., CJK characters) when they appear in the new_string parameter. The last multi-byte character in a string gets truncated at a byte boundary, producing replacement characters (��).

Steps to Reproduce

1. Use the Edit tool to replace text in a file containing CJK (Chinese/Japanese/Korean) characters
2. Include multi-byte UTF-8 characters in the new_string parameter
3. The final multi-byte character in the replacement string may be truncated

Example

Expected new_string content:

**進行中**

Actual result written to file:

**進行��**

The character 中 (U+4E2D, 3-byte UTF-8 sequence E4 B8 AD) was truncated — only the first byte(s) were written, producing the replacement character U+FFFD (��).

Environment

• OS: macOS (Darwin 25.4.0)
• Model: claude-opus-4-6 (1M context)
• Tool: Edit (exact string replacement mode)

Impact

• Silently corrupts file content — no error is raised
• Particularly affects CJK languages where most characters are 3-byte UTF-8 sequences
• Requires manual detection and a follow-up commit to fix