Allow for passing an allocator to the GPU pipeline

OV integration requires temporary buffer allocation. This is to be done using the CPU-side code in the MLIR module. The "kernel" should b e able to accept the allocator as a parameter.

Start by locating the CPU-side code in the MLIR module that handles the GPU pipeline and temporary buffer allocation. Trace how the kernel is defined and invoked; done means the kernel accepts an allocator parameter and uses it for the required temporary buffers.

Written by the indexing model from the issue text.