Fix performance regression of x86_64 GCM with rustc 1.87

[ctz]

rustc 1.87 recently broke inlining of intrinsics where the caller has the proper target_features, but the callee does not. See rust-lang/rust#139029, rust-lang/rust#141309, etc. That looks like (1.86 -> 1.87, main):

aes256-gcm/graviola/8KB time:   [2.2043 µs 2.2170 µs 2.2327 µs]
                        thrpt:  [3.4172 GiB/s 3.4413 GiB/s 3.4612 GiB/s]
                 change:
                        time:   [+44.446% +47.398% +50.394%] (p = 0.00 < 0.05)
                        thrpt:  [-33.508% -32.156% -30.770%]
                        Performance has regressed.

Which is extremely grim. This PR avoids relying on the previous behaviour, and restores performance with 1.87 (main -> this pr):

aes256-gcm/graviola/8KB time:   [1.4685 µs 1.4710 µs 1.4743 µs]
                        thrpt:  [5.1750 GiB/s 5.1866 GiB/s 5.1955 GiB/s]
                 change:
                        time:   [-34.562% -33.443% -32.217%] (p = 0.00 < 0.05)
                        thrpt:  [+47.531% +50.246% +52.818%]
                        Performance has improved.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.31%. Comparing base (08b80ea) to head (9f271e6).
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #90      +/-   ##
==========================================
- Coverage   99.31%   99.31%   -0.01%     
==========================================
  Files         168      168              
  Lines       40269    40265       -4     
==========================================
- Hits        39994    39990       -4     
  Misses        275      275              

☔ 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.

[codspeed-hq]

CodSpeed Performance Report

Merging #90 will improve performances by 10.96%

_{Comparing jbp-x86-gcm-perf-intrinsics (9f271e6) with main (08b80ea)}

Summary

⚡ 2 improvements
✅ 132 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	graviola[8KB]	20.6 µs	18.6 µs	+10.96%
⚡	graviola[8KB]	21.5 µs	19.4 µs	+10.64%