Underflow in `prevfloat` with `Unsigned` iteration count

[enbyd]

prevfloat is implemented by calling nextfloat on the same value but with the iteration count negated:
prevfloat(x::AbstractFloat, d::Integer) = nextfloat(x, -d) julia/base/float.jl#870
For Unsigned iteration counts, this causes underflow and returns unexpected results.

Examples:

prevfloat(1.0, 1)
# 0.9999999999999999        expected result
prevfloat(1.0, 0x01)
# 1.0000000000000566        == nextfloat(1.0, 0xff)
prevfloat(1.0, 0x00000001)
# 1.0000009536743162        == nextfloat(1.0, 0xffffffff)