[CI] Add i8->i32 matmul pack-peel-4-level-tiling performance tests (nod-ai#1072)

Adds new Strix performance tests with 4 levels of tiling, which should
result in better performance. Note that this couldn't be added before as
this needed the fix from
nod-ai#1060.

Author: jtuyls

build_tools/ci/cpu_comparison/run.py

@@ -2082,6 +2082,36 @@ def __init__(self):
                 "run_on_target": "npu4",
                 "skip_numerics": True,
             },
+            {
+                "M": 512,
+                "N": 4096,
+                "K": 512,
+                "in_dtype": "i8",
+                "out_dtype": "i32",
+                "use_ukernel": True,
+                "peano_opt_level": 3,
+                "outline": "all",
+                "transpose_a": False,
+                "transpose_b": False,
+                "tile_pipeline": "pack-peel-4-level-tiling",
+                "run_on_target": "npu4",
+            },
+            {
+                "M": 512,
+                "N": 4096,
+                "K": 512,
+                "in_dtype": "i8",
+                "out_dtype": "i32",
+                "use_ukernel": False,
+                "peano_opt_level": 3,
+                "outline": "all",
+                "outline_to_empty_function": True,
+                "transpose_a": False,
+                "transpose_b": False,
+                "tile_pipeline": "pack-peel-4-level-tiling",
+                "run_on_target": "npu4",
+                "skip_numerics": True,
+            },
         ]
 
         # Some bf16 Performance tests: