fix: prevent panic on empty vectors in AMD64 vector operations

[Fibonacci747]

Fix panic when calling Add/Sub methods on empty vectors in AMD64-optimized
vector implementations across multiple elliptic curve packages.

• Add length check for empty vectors before accessing &a[0] in Add/Sub methods
• Fix InnerProduct validation order to check lengths before empty check
• Ensure consistent behavior with generic implementations
• Apply fixes to all affected curve packages: grumpkin, bn254, stark-curve,
  bls12-381, bls12-377, bls24-315, bls24-317 (both fp and fr packages)
• Add comprehensive test coverage for empty vector operations

This prevents runtime panics that could occur when processing empty vectors
in high-performance cryptographic operations, improving robustness of the
vector math library.

[gbotrel]

Good catch, thanks for the PR.
For your PR to pass the CI, do you mind doing your fix in: vector_ops_asm.go and running go generate ./...?
Also would be nice to have a unit test to accompany it, but can do later 👍

[gbotrel]

I see you already had a commit with tests; similarly for the CI to pass, this needs to go in field/generator/internal/templates/element/tests_vector.go to be generated for all implementations

[Fibonacci747]

I see you already had a commit with tests; similarly for the CI to pass, this needs to go in field/generator/internal/templates/element/tests_vector.go to be generated for all implementations

should i close then this PR and create new PR with changes in correct files? or i should just revert changes in these files to make changes in vector_ops_asm.go and tests_vector.go

[gbotrel]

we do squash merge so you can work on the same PR (go generate ./... in the root will overwrite your changes anyway -- all these files are generated from templates)

[Fibonacci747]

we do squash merge so you can work on the same PR (go generate ./... in the root will overwrite your changes anyway -- all these files are generated from templates)

Done