PoC: compute twiddles on the fly

[Shnatsel]

Computes twiddles on the fly in the variable-size kernel instead of caching them.

This reduces memory usage since the twiddles no longer need to be stored. When combined with #95 we need no auxiliary memory at all for the entire FFT process, including interleaving/deinterleaving and bit reversal.

Zen4 results: on large sizes this is neutral on a single thread and a 12% improvement when multi-threaded. However, small sizes that are not memory-bound do take a large performance penalty, with up to 60% more time taken.

M4 seems to benefit from this on large sizes.

Detailed benchmarks:

Single threaded:

Zen4: https://gist.github.com/Shnatsel/55e32f567c6f1a99ee96f2713bc89b5a
M4: https://gist.github.com/Shnatsel/2695ba6fbc3263a2679f045db5546127

Multi-threaded:

Zen4: https://gist.github.com/Shnatsel/3db0fe24d59c7e4a9a923a1b47653ca0
M4: https://gist.github.com/Shnatsel/e03e1cd148e5208b15448d01223ed5a8

[codecov-commenter]

Codecov Report

❌ Patch coverage is 94.26230% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.54%. Comparing base (3aa61ad) to head (67c97dc).

Files with missing lines	Patch %	Lines
src/kernels/dit.rs	93.26%	7 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #102      +/-   ##
==========================================
- Coverage   98.87%   98.54%   -0.34%     
==========================================
  Files          10       10              
  Lines        2227     2198      -29     
==========================================
- Hits         2202     2166      -36     
- Misses         25       32       +7     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.