Update MKLSparse.jl to support both LP64 and ILP64 interfaces

[amontoison]

No description provided.

[ViralBShah]

Would it be best to try update to the newer API as discussed in #36 (comment).

API: https://www.intel.com/content/www/us/en/docs/onemkl/developer-reference-c/2024-1/blas-and-sparse-blas-routines.html

[ViralBShah]

Rethinking this, perhaps we merge this for now, so that the package at least is non-breaking. Should we put some locks around these calls, so that we don't end up in a bad state if multiple Julia threads are calling into these functions?

I haven't had a chance to investigate the failures.

[amontoison]

@ViralBShah Sorry for not answering earlier.
I finally found time this weekend to work again on MKLSparse.jl.
The fix with this PR is not working.

The best solution is to switch to a newer API.
I started to interface it in 2022 and finalized it today (see PR #33).
Just like MKL.jl, we now support both Int32 and Int64 integers.
This fixes issue #36. I will tag a new release 2.0.0.

I need to check for performance regressions with the new API, but in any case, we must migrate to this new API.