[GPU] dyn_quan bugfix

[isanghao]

Description of the issue(symptom, root-cause, how it was resolved)

• Issue 1) dyn_quan gs128 was supposed to be enabled on xe2+ platforms. It is corrected.
• Issue 2) Proper handling for non-uniform-work-group
  • When kernel gws is not a multiply of lws, it is OK if the platform supports non-uniform-work-group
  • grouped dyn_quan_opt kernel requires such non-uniform-work-group
  • If non-uniform-work-group is not available, disable grouped dynamic quantization
  • gws/lws alignment check is disabled for non-uniform-work-group is supported

Reproduction step and snapshot (if applicable. Do not attach for customer model)

• ./benchmark_app -m faster_rcnn_resnet50_fpn_coco.xml -d GPU.1 --hint latency -niter 1

Problematic graph

Checklist

• Is it a proper fix? (not a workaround) --> it is a proper fix assuming non-uniform-wg is widely available
• Did you include test case for this fix, if necessary?
• Did you review existing test that can be extended to cover this scenario? Which test did you review? --> dyn_quan tests

Tickets:

• 175454