faster? f32 math functions

Attempting to provide faster implementations of common f32 math functions with a similar level of accuracy to the standard library. In the future I want to also provide perfectly rounded versions and less accurate but faster versions. Currently only the cbrt and exp2 functions are significantly faster, but I have a lot of ideas for improvements.

precision (average ulp)

         jodie cbrt error: 0.646
           std cbrt error: 0.094
jodie accurate cbrt error: 0

jodie exp2 error: 1.27
  std exp2 error: 0.985

jodie  cos error: 6.37
  std  cos error: 6.13

jodie  sin error: 9.715
  std  sin error: 7.33

jodie log2 error: 0.055
  std log2 error: 0

benchmarks

Run on i7-12700H

Latency (lower is better)
              | jodie  | std    | perf improvement
--------------|--------|--------|-----------------
         cbrt | 1.4 ns | 7.6 ns | 5.4x
accurate cbrt | 4.0 ns | 7.6 ns | 1.9x
          cos | 2.9 ns | 3.3 ns | 1.1x
         exp2 | 1.0 ns | 2.7 ns | 2.7x
         log2 | 1.0 ns | 1.8 ns | 1.8x
          sin | 2.7 ns | 3.3 ns | 1.2x

Throughput (higher is better) (single thread)
              | jodie    | std      | perf improvement
--------------|----------|----------|-----------------
         cbrt | 5.6 GB/s | 0.5 GB/s | 11.2x
accurate cbrt | 3.8 GB/s | 0.5 GB/s | 7.6x
          cos | 2.5 GB/s | 1.1 GB/s | 2.3x
         exp2 | 6.8 GB/s | 1.4 GB/s | 4.9x
         log2 | 6.6 GB/s | 1.9 GB/s | 3.5x
          sin | 5.0 GB/s | 1.1 GB/s | 4.5x

todo:

• improve cos throughput
• add inverse trig functions
• add tan()
• less accurate but faster versions
• perfectly rounded versions