Float docs do not define the set of "arithmetic" operations

[RalfJung]

Pointed out by @workingjubilee in #129683: the float docs are using the term "arithmetic" operations in a few places, but that term is never defined. Furthermore, the new NaN docs probably should use that term as well when discussing which operations the NaN rules apply to (instead of "non-bitwise floating point operations" which they use currently).

By "arithmetic operations", the docs likely mean all operations except for unary -, abs, copysign, signum, {to,from}_bits, {to,from}_{be,le,ne}_bytes and is_sign_{positive,negative}.