fix(mcp): return valid UTF-8 snippets

[jstar0]

Summary

• Sanitize get_code_snippet source text to valid UTF-8 before serializing the MCP response.
• Preserve valid UTF-8 bytes and replace only invalid source bytes with U+FFFD.
• Add a regression test for non-UTF-8 source bytes in the snippet file.

Fixes #511.

Changes

• Adds a small lossy UTF-8 sanitizer for snippet source strings.
• Uses the sanitized copy for the source field in build_snippet_response().
• Verifies both the raw MCP response envelope and the embedded snippet JSON parse as strict JSON when the source contains CP949/EUC-KR-like bytes.

Verification

make -f Makefile.cbm test
make -f Makefile.cbm lint-format
make -f Makefile.cbm lint-no-suppress
git diff --check

I also attempted:

make -f Makefile.cbm lint-ci

That local run stopped at cppcheck because cppcheck is not installed in my environment.

[DeusData]

Thanks @jstar0 — your reproduce-first test here (write known-bad CP949 bytes, assert the response is valid JSON with U+FFFD replacements) is excellent and we want to keep it.

There's a parallel PR #541 fixing the same issue (#511) with a cleaner sanitizer (canonical (c & 0xF0) == 0xE0 lead-byte checks). Our plan is to land one combined fix: #541's sanitizer + your test + an explicit overflow guard, crediting both of you.

One change needed here regardless: the overflow guard len > (((size_t)-1) - SKIP_ONE) / UTF8_REPLACEMENT_LEN hinges on the opaque SKIP_ONE macro — please make the + 1 explicit (with a comment) so the bound is auditable without chasing the macro.

Would you be open to combining your test with #541's sanitizer? 🙏