Generalize GenerateGhosts for Periodic<Hypercuboid<_>> on N

[janbridley]

Description

Previous code used a fixed list of checks for N=2 and 3 when generating ghosts across a periodic cubic boundary. This change builds up the required list of checks for general dimension, using a bitset to avoid keeping too much data in memory. The result is comparable performance to the original code, but a dimension-agnostic approach. HOWEVER: this changes the ordering of neighbors, which does not appear to have performance or correctness implications but I could be wrong on that.

Motivation and context

HOOMD_MAX_GHOSTS=1 (N match: 1d)
Timer precision: 41 ns
cuboid             fastest       │ slowest       │ median        │ mean          │ samples │ iters
╰─ trait_method                  │               │               │               │         │
   ├─ corner_1d    1.592 ns      │ 1.928 ns      │ 1.612 ns      │ 1.64 ns       │ 100     │ 409600
   ╰─ interior_1d  0.951 ns      │ 4.664 ns      │ 1.002 ns      │ 1.076 ns      │ 100     │ 409600


HOOMD_MAX_GHOSTS=3 (N match: 2d)
Timer precision: 41 ns
cuboid             fastest       │ slowest       │ median        │ mean          │ samples │ iters
├─ specific                      │               │               │               │         │
│  ├─ corner_2d    4.793 ns      │ 25.3 ns       │ 4.875 ns      │ 5.245 ns      │ 100     │ 102400
│  ╰─ interior_2d  3.227 ns      │ 3.695 ns      │ 3.349 ns      │ 3.37 ns       │ 100     │ 204800
╰─ trait_method                  │               │               │               │         │
   ├─ corner_2d    4 ns          │ 4.549 ns      │ 4.142 ns      │ 4.148 ns      │ 100     │ 204800
   ╰─ interior_2d  2.515 ns      │ 3.309 ns      │ 2.616 ns      │ 2.626 ns      │ 100     │ 204800


HOOMD_MAX_GHOSTS=7 (N match: 3d)
Timer precision: 41 ns
cuboid             fastest       │ slowest       │ median        │ mean          │ samples │ iters
├─ specific                      │               │               │               │         │
│  ├─ corner_3d    12.2 ns       │ 13.58 ns      │ 12.69 ns      │ 12.7 ns       │ 100     │ 51200
│  ╰─ interior_3d  7.565 ns      │ 8.704 ns      │ 7.891 ns      │ 7.898 ns      │ 100     │ 51200
╰─ trait_method                  │               │               │               │         │
   ├─ corner_3d    40.67 ns      │ 14.62 µs      │ 40.67 ns      │ 187.2 ns      │ 100     │ 100
   ╰─ interior_3d  7.564 ns      │ 10.08 ns      │ 7.971 ns      │ 7.982 ns      │ 100     │ 102400



HOOMD_MAX_GHOSTS=15 (N match: 4d)
Timer precision: 41 ns
cuboid             fastest       │ slowest       │ median        │ mean          │ samples │ iters
╰─ trait_method                  │               │               │               │         │
   ├─ corner_4d    31.41 ns      │ 38.08 ns      │ 32.71 ns      │ 32.7 ns       │ 100     │ 25600
   ╰─ interior_4d  16.84 ns      │ 19.2 ns       │ 17.41 ns      │ 17.47 ns      │ 100     │ 51200


HOOMD_MAX_GHOSTS=31 (N match: 5d)
Timer precision: 41 ns
cuboid             fastest       │ slowest       │ median        │ mean          │ samples │ iters
╰─ trait_method                  │               │               │               │         │
   ├─ corner_5d    62.18 ns      │ 77.8 ns       │ 64.77 ns      │ 64.97 ns      │ 100     │ 6400
   ╰─ interior_5d  42.97 ns      │ 53.39 ns      │ 43.62 ns      │ 43.84 ns      │ 100     │ 12800

How has this been tested?

Regression tests against the old code

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.