Issue Description:

The random methods on BinaryFloatingPoint are documented with:

/// The `random(in:using:)` static method chooses a random value from a
/// continuous uniform distribution in `range`, and then converts that value
/// to the nearest representable value in this type.

However, the current implementation chooses randomly among 2ⁿ evenly-spaced values in the range, where n is one more than the significand bit-count.

For ranges that span several binades, this coarse-grained comb will miss many representable values in the range.

Furthermore, it will also trap on very large ranges (SR-8798) and it will trap on types where significandBitCount == RawSignificand.bitWidth.

I have written a replacement implementation that solves all of these issues, however I have never contributed significant code to the Swift project before and I would appreciate guidance on how to proceed.