[Feature] Added Sparse Initialization

[cvnad1]

Proposed changes

Adding Feature Sparse Initialization as proposed in the issue #1459

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)

[Saanidhyavats]

@awni, @cvnad1 and I have added sparse init method with documentation and tests. Let us know if any change needs to be done

[angeloskath]

Thanks for the addition!

I think this implementation is unnecessarily complicated. The same can be achieved via a mask. Something as simple as

v = mx.random.normal(shape=a.shape, scale=std, loc=mean, dtype=dtype)
m = mx.random.bernoulli(1-sparsity, shape=a.shape)
return m * v

[zcbenz]

@cvnad1 Sorry for pinging you on an old PR but can you take a look at #3725 to verify whether its claim is correct?