CPU LU factorization and linear solvers

[abeleinin]

Proposed changes

Adds mx.linalg.solve backend following the numpy implementation.

For the LAPACK routines, I chose to use a combination of sgetrf and sgetrs, which is equivalent to calling sgesv (https://stackoverflow.com/a/36065619) because it removes a transpose of a in the C++.

Checklist

Put an x in the boxes that apply.

• I have read the CONTRIBUTING document
• I have run pre-commit run --all-files to format my code / installed pre-commit prior to committing changes
• I have added tests that prove my fix is effective or that my feature works
• I have updated the necessary documentation (if needed)

[awni]

This looks good to me. I'm debating with myself if we should try to do this in high-level ops rather than add a new primitive for it.

On the one-hand it is definitely going to be way less efficient. On the other hand it will run on the GPU and have transform support out of the box.

I'm also ok landing this and rearranging the back-end at a later time if ti makes sense.

[abeleinin]

Were you thinking of an approach using linalg.lu_factor and linalg.solve_triangular instead? Then linalg.solve could just use those ops. I think @barronalex mentioned possibly working on linalg.solve_triangular in a previous review, so maybe those linalg ops should be a higher priority than just linalg.solve support.

[awni]

Then linalg.solve could just use those ops. I think @barronalex mentioned possibly working on linalg.solve_triangular

I think exactly that would be better and more flexible than what we have now. I was even wondering if we could do linalg.solve_triangular using existing MLX ops rather than wrap the lapack routine.

[abeleinin]

Ok, working on adding a LUF primitive and subsequent linalg.lu_factor, linalg.solve_triangular, and linalg.solve functions.

[abeleinin]

Added LUF primitive which adds support for linalg.lu and linalg.lu_factor functions.

Also added two linear solvers, linalg.solve_triangular and linalg.solve using higher-level ops.

This pr closes #30

[sck-at-ucy]

This will be really useful, thank you.