Integrated 64bit long double into the toolchain

[ZERICO2005]

Integrated a 64bit long double type into the toolchain, implemented using Berkeley SoftFloat 3e. SoftFloat is under toolchain/src/softfloat and CEdev/lib/softfloat currently. All CRT long double ABI calls have been implemented, along with some math.h functions.

The following routines are implemented with SoftFloat:

_dadd _dsub _dmul _ddiv _drem
_ftod _ltod _lltod _ultod _ulltod
_dtof _dtol _dtoll _dtoul _dtoull
fmal sqrtl fmodl modfl
truncl floorl ceill roundevenl nearbyintl
roundl lroundl llroundl
rintl lrintl llrint

I have also added the C23 int iszero(real-floating x) function to math.h. I also intend on adding issubnormal, nextup, and nextdown from C23.

[ZERICO2005]

long double float64 appears to work in both CEmu and on the calculator.
Summary of changes since January 11th:

• All the float64 and float32 classification routines have been tested
• float64 arithmetic and conversion works
• optimized some routines in assembly
• libcalls will directly call softfloat functions instead of going through a C wrapper function
• wrote my own versions of _ftod _ltod _ultod _dtol _dtoll _dtoul _dtoull
• ilogbl, frexpl, and ldexpl round correctly and properly handle denormal values
• nextafter, nextup, nextdown, nexttowards are implemented for both float32 and float64
• strtold was added (based off the current strtof routine)
• added -Wimplicit-float-conversion -Wimplicit-int-float-conversion to catch bugs in libc functions
• fixed difftime.c :)
• added C++20 <numbers>
• scalbn now aliases ldexp since FLT_RADIX is 2
• Fixed fmaxf, fminf, and fdimf to handle NaNs/signed-zeros correctly
• added C++ overloads for abs(floaing-point)

Mathematical functions (sinl, expl, tgammal, etc) and remainderl/remquol will be implemented later on.