[LLVMGPU] Use flat workgroup sizes in vector distribution (#18947) #14076

Job	Run time
setup / setup	5s
Build Packages / Linux Release (x86_64)	10m 20s
Test AMD MI250 / test_mi250	5m 8s
Test AMD MI300 / test_mi300	2m 1s
Regression Test / test_regression_suite :: cpu_llvm_task	9m 57s
Regression Test / test_models :: cpu_llvm_task	4m 57s
Regression Test / test_regression_suite :: amdgpu_rocm_mi250_gfx90a	10m 40s
Regression Test / test_models :: amdgpu_rocm_mi250_gfx90a	7m 1s
Test AMD W7900 / test_w7900	1m 49s
Test Android / android_arm64	5m 18s
Regression Test / test_regression_suite :: amdgpu_rocm_mi300_gfx942	8m 34s
Regression Test / test_models :: amdgpu_rocm_mi300_gfx942	3m 29s
Regression Test / test_models :: amdgpu_vulkan	1m 55s
Test ONNX / test_onnx_models :: cpu_llvm_task	3m 9s
Test RISC-V 64 / riscv64	5m 45s
Test ONNX / test_onnx_ops :: cpu_llvm_sync	3m 51s
Unit Test / Linux (x86_64)	3m 1s
Test ONNX / test_onnx_ops :: amdgpu_rocm_rdna3	8m 29s
Test ONNX / test_onnx_ops :: amdgpu_vulkan	1m 42s
Test TensorFlow / Linux (x86_64)	2m 5s
	1h 39m 16s

Provide feedback