feat: Add a pluggable text codec for string patterns - #246

Add StringEncodeDecode: an interface with decode(), encode(), and
encodeLossy(). The host application sets one on the Evaluator. With
none set, a string pattern keeps its raw-byte behavior.

decode() and encode() return std::optional. A nullopt result marks a
byte sequence, or a character, the named encoding cannot represent.
encodeLossy() never fails. It substitutes a replacement character for
anything the encoding cannot represent.

PatternString routes reads and writes through the codec. getValue()
and getBytesOf() throw core::err::E0004 on a nullopt result. A script
can catch this error with try/catch. setValueLossy() writes through
encodeLossy(), and clears the cached bytes and the cached display
string; getBytesOf() otherwise reflects a write only after the next
pattern run.

setValue() clears the cached bytes the same way, on every pattern
type, not just PatternString.

formatDisplayValue() decodes through decode(), throwing the same
core::err::E0004 on a nullopt result. It reads a little past its
display budget, and backs decode() off a few bytes at a time on
failure, so a multi-byte codepoint straddling the read cutoff doesn't
report the whole value as invalid. It then trims the decoded text
itself, not the raw bytes, to the display budget, so a codepoint at
that cutoff stays whole instead of splitting mid-sequence.

getBytesOf() caps the encoded result to the pattern's own size.
getEncodingName() reads the string's own encoding attribute. With
none, it reads the evaluator's default encoding, set through
Evaluator::setDefaultEncoding().

Add a libstd #pragma encoding. It sets the evaluator's default
encoding to its value, unconditionally; the codec, not the pragma,
knows which encoding names are valid.

PatternWideString keeps its fixed UTF-16 behavior. The codec does not
apply to it.

Add PatternLanguage::clearFormatCaches(). It clears every placed
pattern's cached display value, across every section.