doc: float: add note for NaN compare

[inkydragon]

Current documentation seems to imply that one should use === for NaN comparisons.
But non-normalized NaNs will give unexpected results: NaN32p1 !== NaN32.

julia> reinterpret(UInt32, NaN32)
0x7fc00000

julia> NaN32p1 = reinterpret(Float32, 0x7fc00001)
NaN32

julia> NaN32p1 === NaN32,  NaN32p1 == NaN,  isequal(NaN32p1, NaN32),  isnan(NaN32p1)
(false, false, true, true)

julia> NaN32 === NaN32,  NaN32 == NaN,  isequal(NaN32, NaN32),  isnan(NaN32)
(true, false, true, true)

Maybe a little relevant (NaN canonicalization is mentioned): #49353

[mikmoore]

If we're going to remark on the unsuitability of ===, I think we should provide an explanation. Maybe modify the remark on === to something vaguely like the following?

"There are many different encodings of NaN. Do not rely on === for NaN comparisons because isnan(a) & isnan(b) does not imply a === b."

This might enable a slightly simpler (is it?) example like

julia> a, b = reinterpret(Float64, Int64(-1)), reinterpret(Float64, Int64(-2))
(NaN, NaN)

julia> isequal(a, b), a === b
(true, false)