Denormal floating-point constants

[HertzDevil]

Float64::MIN_POSITIVE is the "Smallest representable positive value" for Float64, and similarly for Float32:

struct Float64
  # Smallest representable positive value
  MIN_POSITIVE = 2.2250738585072014e-308_f64
end

But this is inaccurate because MIN_POSITIVE is the smallest normal floating-point value; there are 2 ** 52 - 1 denormal floats smaller than that, and all of them do represent exact values, the smallest of which admits any of the following equivalent definitions:

5.0e-324_f64
0.0_f64.next_float
1_u64.unsafe_as(Float64)

Or, for Float32:

1.0e-45_f32
0.0_f32.next_float
1_u32.unsafe_as(Float32)

I suggest we add these constants as MIN_DENORMAL, and then adjust MIN_POSITIVE's docs accordingly.