[AMD][GLUON] Support scalar other value for loads (#8039)

Support scalar other value for following ops:

- `amd.cdna3.buffer_load`
- `amd.cdna4.async_copy.buffer_load_to_shared` 
- `amd.cdna4.async_copy.global_load_to_shared`

We will explicitly convert scalar other to a tensor and cast to match
the pointer type.

Author: borontion