gpu: intel: compute: fix large GRF mode query with ARL-H

The workaround is no more relevant, DG2 A0 is used neither in
development nor in production. It is also more reliable and correct to
assume that systolic support implies large GRF mode support.

src/gpu/intel/compute/compute_engine.hpp

@@ -163,12 +163,7 @@ class compute_engine_t : public engine_t {
                 : device_info_->gpu_arch() >= compute::gpu_arch_t::xe_hpc;
     }
     bool mayiuse_large_grf_mode() const {
-        // XXX: XeHPG 128EU A0 causes hangs with large GRF mode.
-        if (is_xe_hpg() && device_info()->eu_count() == 128
-                && device_info()->stepping_id() == 0
-                && device_info()->mayiuse_systolic())
-            return false;
-        return device_info_->gpu_arch() >= compute::gpu_arch_t::xe_hp;
+        return device_info()->mayiuse_systolic();
     }
 
     dispatch_t create_dispatch(const memory_desc_t *md = nullptr) const {