This allows us to catch the case where a match
occurs without optimizations, but doesn't occur
with optimizations. Additionally fix the `xfail`
param such that it can't be used on tests that
actually match expectations.

Replace a couple of `#if os(Linux)` checks with
a check to see if we have a newer stdlib
available. This lets us emit an expected failure
in the case where we're testing on an older
stdlib.

Previously we performed a lexicographic
comparison with the bounds of a character class
range. However this produced surprising results,
and our implementation didn't properly handle
case sensitivity.

Update the logic to instead only allow single
scalar NFC bounds. The input is then converted to
NFC in grapheme semantic mode, and checked against
the range. In scalar semantic mode, the input
scalar is checked on its own. Additionally, fix
the case sensitivity handling such that we check
both the lowercase and uppercase version of the
input against the range.

Previously we would emit a series of scalars
written in the DSL as a series of individual
characters in grapheme semantic mode. Change the
behavior such that we coalesce any adjacent
scalars and characters, including those in regex
literals and nested concatenations. We then
perform grapheme breaking over the result, and can
emit character matches for scalars that coalesced
into a grapheme.

This transform subsumes a similar transform we
performed for regex literals when converting them
to a DSLTree. This has the nice side effect of
allowing us to better preserve scalar syntax in
the DSL transform.

rdar://96942688

Previously we would only match entire characters.
Update to use the generic Character consumer logic
that can handle scalar semantic mode.

rdar://97209131

In grapheme semantic mode, coalesce adjacent
character and scalar members of a custom character
class, over which we can perform grapheme breaking.
This involves potentially re-writing ranges such
that they contain a complete grapheme of adjacent
scalars.

Make sure we throw the right error for ranges
that are invalid in grapheme mode, but are valid
in scalar mode.

I also noticed that `lexQuantifier` could silently
eat trivia if it failed to lex a quantification,
so also fix that.