string.c: bound check mrb_str_byte_to_char before use - #7098

Follow-up to #7097, which promoted chars2bytes / bytes2chars to
mrb_str_char_to_byte / mrb_str_byte_to_char. This addresses a review
comment that arrived after the merge.

mrb_str_byte_to_char formed p + bi before testing the offset, so a byte
offset outside the string produced an out-of-bounds pointer, and the
single-byte and binary fast path returned that offset unchanged instead of
-1. Moving the bounds test to the top covers the fast path and the pointer
arithmetic alike, and the e < pivot test it replaces goes away.

The non-UTF-8 identity function gets the same test, so the documented
contract holds on every build: -1 when the offset lies outside the string,
and on UTF-8 builds also when it lands inside a multi-byte character. Without
this, the same call would answer differently depending on MRB_UTF8_STRING,
which is the opposite of what promoting these functions was for.

Every caller clamps the offset it passes today (str_rindex results,
str_index_str_by_char, and re_byte_to_char in mruby-regexp, which clamps
to RSTRING_LEN before calling), so no behavior changes. The check belongs
in the function now that it is internal API rather than a static confined to
string.c.

Testing

• rake test with build_config/host-debug.rb (full-core, MRB_UTF8_STRING):
  mrbtest 2233 OK / 0 KO, bintest 116 OK.
• rake test with the default config (byte strings, identity conversion):
  mrbtest 2032 OK / 0 KO, bintest 105 OK.

Summary by CodeRabbit

• Bug Fixes

  • String byte-to-character conversion now rejects negative and out-of-range byte offsets consistently.
  • Invalid offsets return an error instead of producing unexpected character positions.
  • Non-UTF-8 conversions continue to behave as identity operations while enforcing valid bounds.

• Documentation

  • Clarified the accepted offset range and behavior for non-UTF-8 strings.