[ET-VK][ez] Test specific sizes of linear sizes in generated operator tests (pytorch#7672)

Pull Request resolved: pytorch#7667

## Context

Recent changes related to checking SPIR-V capability support at runtime have made it possible to test the 8-bit quantized linear compute shader on Android devices. Previously the test would be automatically skipped since the operator potentially uses 8-bit data types.

To make the generated tests more useful, instead test real sizes of linear layer settings found in a sample model in the 8-bit linear test case.
ghstack-source-id: 261524380
@exported-using-ghexport

Differential Revision: [D68192068](https://our.internmc.facebook.com/intern/diff/D68192068/)

Co-authored-by: Stephen Jia <[email protected]>

Author: pytorchbot

backends/vulkan/test/op_tests/cases.py

@@ -169,7 +169,13 @@ def get_linear_test_suites():
 
 @register_test_suite("aten._weight_int8pack_mm.default")
 def get_weight_int8pack_mm_inputs():
-    MKN_list = common_MKN_list
+    MKN_list = [
+        [6, 480, 256],
+        [6, 256, 1024],
+        [6, 1024, 256],
+        [6, 256, 256],
+        [6, 256, 512],
+    ]
 
     inputs_list = [((M, K), (N, K), (N)) for M, K, N in MKN_list]