This PR extends index_fill and index_copy operations to support int32 indices in addition to the existing int64 support:

1. Memory Efficiency and potential performance for handling a large number of indices, particularly when needing to transfer to accelerator backends. In some cases, the compiler may not support 64-bit integers, and in some cases, may cause conflicting casts when used with TorchXLA. I do not see the immediate need to propagate these to the dim input, since it is tied to the APIs and is relatively negligible.
2. Framework interoperability: As mentioned above, this gives more flexibility when working with TorchXLA, since some operations require the same type of physical raw representations for the tensors as the XLA tensors. In some cases, for Neuron, XLA generates S32 types which are not compatible with some operations (e.g. casts) when needing to convert across tensors.
3. Consistency with other APIs, such as index_add and index_select.

cc @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10