Leverage TE wgrad accumulation fusion in MFSDP v2

MFSDP v2 currently materializes local wgrads, then copies them into the reduce-scatter input buffer.

In `experimental/module.py::_reduce_gradient_groups()`, MFSDP v2 allocates the partial-gradient buffer and calls `copy_gradients_to_partial_buffer()`. `experimental/parameter_group.py::copy_gradients_to_partial_buffer()` then copies each `unsharded.grad` into the corresponding partial-buffer view before reduce-scatter.

Transformer Engine wgrad accumulation fusion can write wgrads directly into a preallocated destination buffer. Using the MFSDP v2 reduce-scatter input views as that destination should eliminate this extra full-gradient copy, as well as the separate transient autograd-gradient buffer.

TE enables this through its wgrad-accumulation option and a preallocated `main_grad`-style destination.

**Done when:** supported TE linear layers write directly to MFSDP v2’s reduce-scatter input buffer; gradients match the existing path; and a benchmark reports memory and step-time impact.

Read experimental/module.py::_reduce_gradient_groups() and experimental/parameter_group.py::copy_gradients_to_partial_buffer() first, then trace the supported Transformer Engine linear-layer path and its wgrad-accumulation option. Verify gradients against the existing path and run a benchmark reporting memory and step-time impact; done means supported TE layers write directly to the MFSDP v2 reduce-scatter input buffer.

Written by the indexing model from the issue text.