Report UTF-8 bytes in MCP write tool

[fallintoplace]

Problem

The agent write tool returns a status string with len(content), which is a character count, while write operations are byte-oriented.

For non-ASCII text this reports a smaller number than the actual persisted bytes, e.g. "é".

Fix

• In src/anthropic/lib/tools/agent_toolset.py, update the return message to use UTF-8 encoded byte length:
  • len(content.encode("utf-8"))

This keeps the current bytes status wording accurate while preserving behavior of writing with UTF-8 (the same default used by Path.write_text).

Testing

Not run (docs/runtime behavior change only).