Give put_along_axis axis a default of None like take_along_axis

[ayaangazali]

What is wrong

mx.put_along_axis documents axis as optional with a default of None, and its implementation already handles the None case (flatten to 1D then index), but the binding marks axis as required. Calling it without axis raises, and the error message prints the very signature it just rejected:

import mlx.core as mx
a = mx.array([[1, 2], [3, 4]])
mx.put_along_axis(a, mx.array([0, 3]), mx.array([7, 8]))

TypeError: put_along_axis(): incompatible function arguments. The following argument types are supported:
    1. put_along_axis(a: array, /, indices: array, values: array, axis: int | None = None, *, stream: StreamOrDevice = None) -> array

The nb::sig string and the docstring both say axis: int | None = None, so the reference and the actual binding disagree.

Why

The parameter is declared "axis"_a.none(), which permits None to be passed but sets no default value, so the argument stays required. This is the same defect that #4357 just fixed for the sibling take_along_axis; put_along_axis has the identical signature and the identical axis.has_value() flatten path, and was missed.

What this changes

One line, mirroring #4357:

-      "axis"_a.none(),
+      "axis"_a = nb::none(),

After it, the flattened form works and matches the documented signature:

mx.put_along_axis(a, mx.array([0, 3]), mx.array([7, 8]))          # [[7, 2], [3, 8]]
mx.put_along_axis(a, mx.array([0, 3]), mx.array([7, 8]), axis=None)  # same
mx.put_along_axis(a, mx.array([[0],[1]]), mx.array([[9],[9]]), axis=1)  # unchanged

No test added, matching #4357 which shipped the same one-line fix for take_along_axis without one. Full CPU rebuild, clang-format clean, and the compiled binding verified by the calls above.

---

I am new and lean on Claude Code, so I kept this to the exact shape of the merged sibling fix and checked the rebuilt binding by hand rather than trusting the source. Point me at anything you want done differently.