Numba Reduction operations (+=, *=) not supported on GPU

[adarshyoga]

Add-assign, multiply-assign operations within Numba par-fors which are reduction operations need to be supported on GPU.

Here is an example kernel in pygbm where "+=" reduction operation is necessary.

@njit(parallel=True)
def _update_raw_predictions(leaves_data, raw_predictions):
  for leaf_idx in prange(len(leaves_data)):
     leaf_value, sample_indices = leaves_data[leaf_idx]
     for sample_idx in sample_indices:
         raw_predictions[sample_idx] += leaf_value

[diptorupd]

TODOs to get prange reductions for + and * working in main:

• Update Enable dpnp parfor reduction #999 on top of current main
• Fix SLM constant size validation pass
• Turn on block renaming and remove dels options in gufunc kernel generation
• Review Enable dpnp parfor reduction #999 and do any further refactoring as needed (code quality)
• Add unit tests

[diptorupd]

Fixed in #999