fix(config): tolerate UTF-8 BOM in config files

[Jerry2003826]

Fixes #2043

Summary

• Read disk config files with utf-8-sig so a leading UTF-8 BOM is ignored before TOML/JSON parsing.
• Add regression coverage for a BOM-prefixed config.toml.
• Keep config schema and save behavior unchanged.

Root Cause

load_config() read config files with utf-8, which preserves a leading \ufeff. tomlkit then parsed that BOM as TOML content and raised Empty key at line 1 col 0.

Validation

• uv run pytest tests/core/test_config.py -k "bom" -q
• uv run pytest tests/core/test_config.py -q
• uv run ruff check src/kimi_cli/config.py tests/core/test_config.py
• uv run ruff format --check src/kimi_cli/config.py tests/core/test_config.py
• uv run pyright src/kimi_cli/config.py tests/core/test_config.py

Note

• On this Windows machine, full uv run pyright still reports existing POSIX-only attribute issues in tests/e2e/shell_pty_helpers.py; the changed files pass targeted pyright.