Switch to back to libblas's BLAS implementation

[Songtech-0912]

Currently we use the matrixmultiply crate for providing BLAS operations such as matrix multiplication. This is fine but matrixmultiply uses unsafe code and also it only implements matrix multiplication which is not ideal for our purposes.

Instead, it would be best to switch back to libblas (it's already in our Cargo.toml, we're just not using it), which is safe, implements level 1-3 operations, and is fast. We don't need more unsafe in elara-array than we absolutely need for performance/functionality.