doc: say what a build converting case by ASCII stops refusing - #7231

doc/guides/mrbconf.md, under MRB_UTF8_STRING:

• A string read as bytes (String#b) converts and folds ASCII alone, and one
  holding bytes that spell no character is refused with ArgumentError.

and, two bullets down, what the option beside it takes away:

• MRB_USE_ASCII_CASE narrows the case half back to ASCII, leaving the
  indexing.

The MRB_USE_ASCII_CASE section says the same at greater length: it narrows which
characters convert, and names the case table and the regexp i flag as what goes with
them. Neither place says the refusal goes too, so a reader carries the promise across.

What happens

It goes too. The refusal is raised inside the walk over characters, src/string.c:2197,
and the whole walk sits inside #if defined(MRB_UTF8_STRING) && !defined(MRB_USE_ASCII_CASE),
src/string.c:2080-2251. Compiled out, every method falls to its own ASCII loop, and
that loop asks nothing about what the bytes spell.

"abc\x80", on three of the build_config/ci/gcc-clang.rb builds:

Every ArgumentError above carries the message input string invalid.

Why this changes the documents and not the code

The refusal is there because answering as though a stray byte were the character it
starts with hands back a string nobody asked for. A build that converts nothing above
ASCII writes nothing over those bytes, so there is no wrong answer left to refuse, and
reading the string through to find them would buy only the exception. The ascii-case
column above is also the byte-string column, which is the shape the last bullet of the
section already promises: a build reading its strings as bytes converts ASCII alone
whatever the option says.

Both readings are pinned already:

• test/t/string.rb:452 asks for the ArgumentError from #downcase, #upcase,
  #capitalize and #downcase!, gated if UNICODECASE.
• test/t/string.rb:498 asks the same bytes back untouched from #downcase and
  #upcase, gated unless UNICODECASE.
• mrbgems/mruby-string-ext/test/string.rb:360-362 asks for it from #casecmp? and
  #swapcase, gated skip unless UNICODECASE, and :376 asks #casecmp? for false
  instead, gated skip if UNICODECASE.

So what is missing is the sentence.

Fix

Three sentences, no code.

• The MRB_USE_ASCII_CASE section gains a bullet: bytes that spell no character are
  handed back as they stand rather than refused, the refusal belonging to the walk the
  option narrows away.
• The MRB_UTF8_STRING bullet that sends the reader to that option names the refusal
  as part of what it takes, so the promise two bullets above is not carried across.
• doc/limitations.md, which said only that the option "narrows that half back without
  giving up the indexing", gains the same qualification.

Verification

No build output changes: the diff is two Markdown files. The table above was measured on
this branch, and rake -m test on build_config/ci/gcc-clang.rb reports KO 0 and
Crash 0 in all five builds, unchanged from 03daa666.

Environment

# ci/gcc-clang, bintest
gcc -std=gnu99 -g -O3 -Wall -Wundef -Werror-implicit-function-declaration -Wwrite-strings -DMRB_GC_FIXED_ARENA -DMRB_USE_BIGINT -DMRB_USE_COMPLEX -DHAVE_MRUBY_ENCODING_GEM -DMRB_UTF8_STRING -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DMRB_USE_TASK_SCHEDULER -DMRB_USE_DEBUG_HOOK src/string.c

# ci/gcc-clang, ascii-case
gcc -std=gnu99 -g -O3 -Wall -Wundef -Werror-implicit-function-declaration -Wwrite-strings -DMRB_USE_ASCII_CASE -DMRB_USE_BIGINT -DMRB_USE_COMPLEX -DHAVE_MRUBY_ENCODING_GEM -DMRB_UTF8_STRING -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DMRB_USE_TASK_SCHEDULER src/string.c

# ci/gcc-clang, byte-string
gcc -std=gnu99 -g -O3 -Wall -Wundef -Werror-implicit-function-declaration -Wwrite-strings -DMRB_USE_BIGINT -DMRB_USE_COMPLEX -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DMRB_USE_TASK_SCHEDULER src/string.c

🤖 Generated with Claude Code

https://claude.ai/code/session_01F57Qh8PB4NJRGZPbg9QvHU

Summary by CodeRabbit

• Documentation
  • Clarified that ASCII narrowing returns invalid UTF-8 byte sequences unchanged instead of raising an ArgumentError.
  • Expanded encoding limitation guidance to explain behavior under Unicode conversion and ASCII narrowing.