Fix constant case losing idempotency around digits

[afonsojanu]

constantCase (and any of the other helpers that route through the same word
splitter, e.g. capitalCase, snakeCase, dotCase...) can lose its own
output the second time it runs.

constantCase("foo123bar");                    // "FOO123BAR"
constantCase(constantCase("foo123bar"));       // "FOO123_BAR"  <- extra "_"

constantCase("a1b2c3");                        // "A1B2C3"
constantCase(constantCase("a1b2c3"));          // "A1_B2_C3"    <- extra "_"s

The splitter's SPLIT_LOWER_UPPER_RE treats any digit directly followed by
an upper case letter as a word break, regardless of what's before the digit.
That's fine the first time (there's no upper case letter in foo123bar to
begin with), but once a word has been shouted into all caps, the same digit
is now followed by an upper case letter purely as an artifact of that
shouting, not because a new word actually started there, so re-splitting
inserts a bogus boundary.

I confirmed this against the reported repro and then fuzzed constantCase
against 200k random ASCII strings comparing f(x) to f(f(x)); the
unmodified splitter fails on roughly 4% of them, all following this same
digit pattern.

The fix keeps the "digit ends a word" rule, but only when the following
upper case letter is itself the start of a real word, i.e. followed by a
lower case letter (like the T in V1Test). An already-shouted run such as
FOO123BAR no longer matches, since the letter after the digit isn't
followed by a lower case letter anywhere nearby, it's followed by more
capitals.

With that change the same 200k-iteration fuzz run comes back with zero
constantCase mismatches, the full existing suite still passes (141 tests
across the monorepo, including the V1Test, TestV2 and Foo12019Bar
cases that specifically exercise digit/case boundaries), and I added a
regression test covering the two examples above.

[afonsojanu]

Closing this for now as part of a clean slate — stepping back from everything I had open across these repos.