Clarification for max work group size in kernel

[aravindksg]

Depending on kernel properties, the number of threads per thread group might be different from what we have in ze_device_compute_properties.maxTotalGroupSize . For example, large GRF mode would require half number of threads per thread group to be programed.

Ask here is to either introduce a new kernel property that returns a value that is less than or equal to the maxTotalGroupSize device property (this would be similar to CL_KERNEL_WORK_GROUP_SIZE) or clarify that for large GRF mode, half number of threads per thread group will be used.