convert all for_each to for loops in DiT kernels

[Shnatsel]

required for runtime feature detection on x86 because closures break up the chain of #[inline(always)]

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.79%. Comparing base (dd00bef) to head (426f9db).

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #78   +/-   ##
=======================================
  Coverage   99.79%   99.79%           
=======================================
  Files          13       13           
  Lines        2876     2965   +89     
=======================================
+ Hits         2870     2959   +89     
  Misses          6        6           

☔ 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.

[Shnatsel]

This causes a performance regression on ARM, so this cannot be merged as-is

[Shnatsel]

closing in favor of #79

[LaurenzV]

Can you not just annotate the closures themselves with inline always?

[Shnatsel]

That doesn't do anything because each for_each is a function that doesn't get inlined. I've tried.