Code cleanups: Unify _Float_traits and _Floating_type_traits

[StephanTLavavej]

• tr1: _HAS_WINDOWS_FILESYSTEM is unused.
  • We don't usually clean up tr1, but this was polluting my search for #ifdef _HAS.
• cond.cpp, mutex.cpp: Use alignof.
• Hyphenate floating-point in comments.
• Fix "non meow" comments (usually to say "non-meow", sometimes rephrasing differently).
• Replace _Float_traits usage with _Floating_type_traits from <type_traits>.
  • While @statementreply was working on <complex>: Improve numerical accuracy of sqrt and log #935 which added _Float_traits to <xutility>, Finish P0768R1 by adding Spaceship CPOs #1370 moved _Floating_type_traits up from <charconv> to <type_traits>. We should de-duplicate this machinery now. _Floating_type_traits has all of this content and more (under different names), except for _Float_traits::_Magnitude_mask. This is used only once in product code (_Float_abs_bits), where I believe it's simpler to refer to ~_Floating_type_traits::_Shifted_sign_mask (as we're clearly masking away the sign bit), instead of adding _Magnitude_mask to _Floating_type_traits.
  • I verified with static_assert that the values are identical.
  • Note that there is a terminology difference: _Floating_type_traits has _Shifted_exponent_mask (which we want; it's been "shifted" into the proper position for single-precision or double-precision) and _Exponent_mask (which we do not want). I double-checked that all uses of _Float_traits::_Exponent_mask were properly converted to _Floating_type_traits::_Shifted_exponent_mask.
  • This extracts _Traits and _Uint_type typedefs for readability.
  • This also updates the test code's definition of magnitude_mask_v; the constants need to be re-ordered which is why the diff looks complicated.