feat: Add AdamW bias correction parameter.

[mattgallagher]

Proposed changes

This PR adds a biasCorrection parameter to the Swift Adam and AdamW optimizers, matching the Python API and bias-corrected update rule used in MLX Python.

The main motivation is consistency with the Python implementation and easier comparison against PyTorch and other common optimizer implementations. I don't know if this falls within the goals of the mlx-swift project but I definitely encountered a situation where this was preventing close comparison with an MPSGraph implementation which was itself written to behave like PyTorch.

Internally, this requires tracking the optimizer step so the bias-correction factors can be applied:

• 1 - beta1^t
• 1 - beta2^t

To support that, this change replaces TupleState with an AdamState for Adam/AdamW, so the first moment, second moment, and step can be stored together.

This PR also adds tests for the bias-corrected Adam and AdamW paths and updates the relevant optimizer documentation.

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)

[davidkoski]

Looks good, thanks!