[libc++] LWG2381: Inconsistency in parsing floating point numbers (#77948)

This PR implements [LWG2381](https://cplusplus.github.io/LWG/issue2381)
by rejecting `'i'`, `'I'`, `'n'`, `'N'` in FP parsing, as inf and NaN
are intendedly rejected by that LWG issue.

The source character array used for parsing is
`"0123456789abcdefABCDEFxX+-pPiInN"`, whose first 26 or 28 characters
are used for parsing integers or floating-point values respectively.
Previously, libc++ used 32 characters, including `'i'`, `'I'`, `'n'`,
`'N'`, for FP parsing, which was inconsistent with LWG2381. This PR also
replaces magic numbers 26 and 28 (formerly 32) with named constants.

Drive-by change: when the first character (possibly after the leading
`'+'` or `'-'`) is not a decimal digit but an acceptable character
(e.g., `'p'` or `'e'`), the character is not accumulated now (per Stage
2 in [facet.num.get.virtuals]/3).

#65168 may be rendered invalid, see
https://github.com/llvm/llvm-project/pull/65168#issuecomment-1868533342

.

Apple back-deployment targets remain broken, likely due to dylib. XFAIL
is marked in related tests.

---------

Co-authored-by: Mark de Wever <koraq@xs4all.nl>