Replaces each maximal run of invalid UTF-8 bytes with a replacement
string (U+FFFD by default), returning a valid UTF-8 copy. Mirrors
CRuby's String#scrub (Feature #6752) -- the recovery counterpart to
the existing String#valid_encoding? detection API.
Validation matches utf8code() in src/string.c after the RFC 3629 /
Unicode D93b conformance fixup (#2708): overlong encodings, UTF-16
surrogates, and codepoints above U+10FFFF are all treated as invalid.
This is stricter than the existing mrb_utf8len()-based check used by
valid_encoding?, so a string can report valid_encoding? = true and
still get scrubbed; aligning valid_encoding? is a follow-up.
The block form lives in mrblib on top of two C primitives -- __scrub
and __scrub_chunks -- to avoid VM re-entry from C per CLAUDE.md.
Non-String block return values are coerced via to_s (CRuby raises
TypeError instead; the choice is locked in by test).
Closes #6859.
Co-authored-by: Claude <noreply@anthropic.com>