[samkim-crypto]

Summary of Changes

This is the final PR before we kick off the audits.

When I tested the crate with SBF v3, the CUs increased a bit, so I added two more optimizations:

• Scalar reduction: I saw that we zeroed the limb array before filling it in a loop. The SBF v3 measurements showed that this setup and decoding were costly, so I changed it to initialize every limb directly, simplified carry propagation, and only apply the final correction when needed.
• Point validation: I saw that strict verification made extra curve calls for small-order checks before the MSM. Those calls added CUs even though the verification operations already enforce curve validity, so I switched to checks on the compressed encodings and reused the MSM and fallback operations for validation.

[zz-sol]

LGTM