Run streflop-float-test on x86_64 and arm64 runners

[domenix]

Restructured the streflop-float-test workflow to run on both x86_64 and arm64 architectures and perform an automated comparison of the results

I modified it to only compare the hashes not the full binaries since it added large overhead to the workflow, which now would run as part of PRs touching relevant code

I left the python script in as reference if binary comparison is needed in the future

[domenix]

Workflow was tested and ran with success:
https://github.com/domenix/RecoilEngine/actions/runs/22980804175

Comparison output:
=== 52K tests ===
Category A: MATCH (028748A131B6DE45)
Category B: MATCH (90A499DDF050A110)
Category C: MATCH (8AC013D704C7DC9C)
Category D: MATCH (AD402488FCDFB689)
Category E: DIFFERS (x86=29A2D3322E8FC9B2 arm=FA1C908925CE5EA8) [informational — edge/NaN cases]
PASS: categories A-D are identical
=== 261M tests ===
Category A: MATCH (3D7BC3F410005848)
Category B: MATCH (6EE54BD05C7EF756)
Category C: MATCH (45E501F69C1EF72D)
Category D: MATCH (DFB5415E5C9DD4F7)
Category E: DIFFERS (x86=4DA9084609AF0DCD arm=5B7233B5F7B5AEB3) [informational — edge/NaN cases]
PASS: categories A-D are identical
=== 2.6B tests ===
Category A: MATCH (B1CB906A23A4ECC6)
Category B: MATCH (235DADDCFB1ACC0E)
Category C: MATCH (B95CBEE109FFD162)
Category D: MATCH (E4CA2FEBEE2E2066)
Category E: DIFFERS (x86=92940180E56CA9A2 arm=42C9BEA82C138B48) [informational — edge/NaN cases]
PASS: categories A-D are identical