Add dummy Linalg ops to SSVE tests (#14099)

The lowering pipelines are not really prepared to deal with empty
dispatches. This is unrelated to the logic in the tests for SSVE
attributes (which, atm, have empty dispatches), so I'm adding some dummy
Linalg ops to workaround it.

compiler/src/iree/compiler/Codegen/LLVMCPU/test/pipeline_tests.mlir

diff --git a/compiler/src/iree/compiler/Codegen/LLVMCPU/test/pipeline_tests.mlir b/compiler/src/iree/compiler/Codegen/LLVMCPU/test/pipeline_tests.mlir
index 588d886..a95fa96 100644
--- a/compiler/src/iree/compiler/Codegen/LLVMCPU/test/pipeline_tests.mlir
+++ b/compiler/src/iree/compiler/Codegen/LLVMCPU/test/pipeline_tests.mlir
@@ -634,7 +634,12 @@
       hal.return %arg1, %arg2, %arg2 : index, index, index
     }
     builtin.module {
-      func.func @dispatch() { return }
+      func.func @dispatch() { 
+        %cst_0 = arith.constant 0.000000e+00 : f32
+        %0 = tensor.empty() : tensor<1xf32>
+        %1 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<1xf32>) -> tensor<1xf32>
+        return
+      }
     }
   }
 }
@@ -651,7 +656,12 @@
       hal.return %arg1, %arg2, %arg2 : index, index, index
     }
     builtin.module {
-      func.func @dispatch() { return }
+      func.func @dispatch() { 
+        %cst_0 = arith.constant 0.000000e+00 : f32
+        %0 = tensor.empty() : tensor<1xf32>
+        %1 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<1xf32>) -> tensor<1xf32>
+        return
+      }
     }
   }
 }
@@ -668,7 +678,12 @@
       hal.return %arg1, %arg2, %arg2 : index, index, index
     }
     builtin.module {
-      func.func @dispatch() { return }
+      func.func @dispatch() {
+        %cst_0 = arith.constant 0.000000e+00 : f32
+        %0 = tensor.empty() : tensor<1xf32>
+        %1 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<1xf32>) -> tensor<1xf32>
+        return
+      }
     }
   }
 }
@@ -695,7 +710,12 @@
       hal.return %arg1, %arg2, %arg2 : index, index, index
     }
     builtin.module {
-      func.func @dispatch() { return }
+      func.func @dispatch() {
+        %cst_0 = arith.constant 0.000000e+00 : f32
+        %0 = tensor.empty() : tensor<1xf32>
+        %1 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<1xf32>) -> tensor<1xf32>
+        return
+      }
     }
   }
 }