Test out single-precision support mapping

[janbridley]

Description

Improve performance of composite support mapping.

NOTE: this code is not of a sufficient quality to be incorporated, and the finding needs significant further review -- however, the performance delta running SupportMapping for ConvexPolyhedron in single precision is almost 50% on osx-arm64.

Motivation and context

# f64

> ./target/release/benchmark -b mc_3d_octahedron
[INFO  benchmark] mc_3d_octahedron: 4096 bodies
[INFO  benchmark] 1 threads
[INFO  benchmarks::benchmark] Average: 90.90262056771684 sweep/s
mc_3d_octahedron    :    90.903 steps/s

# f32

> ./target/release/benchmark -b mc_3d_octahedron
[INFO  benchmark] mc_3d_octahedron: 4096 bodies
[INFO  benchmark] 1 threads
[INFO  benchmarks::benchmark] Average: 139.90105206632083 sweep/s
mc_3d_octahedron    :   139.901 steps/s

How has this been tested?

Checklist:

• I have reviewed the Contributor Guidelines.
• I agree with the terms of the hoomd-rs Contributor Agreement.
• My name is on the list of contributors (doc/src/credits.md) in the pull request source branch.
• I have summarized these changes in release-notes.md following the established format.