Fix TestExpRotation1BlockOfFour on arm64

[thomas-vilte]

Description

TestExpRotation1BlockOfFour compares the unrolled expRotation1 with a scalar reference using assert.Equal. On arm64 the compiler contracts the multiply-adds into FMA, and it does so differently for the two versions, so they drift in the last float32 bit (-6.7551584 vs -6.755159). The test only passed on amd64, and on the macOS runner, because that job runs with -race, which changes codegen enough to hide it.

The rotation itself is fine, so this only changes the test, to InDeltaSlice with a 1e-5 tolerance. The largest difference I saw was about 6e-7 on values up to ~10, so this leaves room for a few ulps while still catching a real mistake in the unrolled path.

With this change, go test ./... passes on arm64 without -race (qemu, Go 1.27). Before, this was the only failing test. amd64 passes with and without -race.

Reference issue

Fixes #241

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.11%. Comparing base (9c08810) to head (42d1881).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #252   +/-   ##
=======================================
  Coverage   94.11%   94.11%           
=======================================
  Files          63       63           
  Lines       11119    11119           
=======================================
  Hits        10465    10465           
  Misses        453      453           
  Partials      201      201           

Flag	Coverage Δ
go	94.11% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 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.