[mlir][linalg] Fix numerical issue with softmax (#96090)

For more info:
iree-org/iree#17670 (comment)

Author: pashu123

mlir/lib/Dialect/Linalg/IR/LinalgOps.cpp

@@ -2714,8 +2714,8 @@ FailureOr<SmallVector<Value>> SoftmaxOp::decomposeOperation(OpBuilder &b) {
   Value neutralForMaxFInit =
       b.create<linalg::FillOp>(loc, Value{neutralForMaxF}, outputReduce)
           .result();
-  Value max = reduce<arith::MaximumFOp>(b, loc, input, neutralForMaxFInit,
-                                        reductionDim);
+  Value max =
+      reduce<arith::MaxNumFOp>(b, loc, input, neutralForMaxFInit, reductionDim);
 
   // Step 2: Subtract max from input and exponentiate.
   Value numerator = buildSubAndExpOp(b, loc, input, max, output, reductionDim);

mlir/test/Dialect/Linalg/transform-op-decompose.mlir

@@ -215,7 +215,7 @@ func.func @softmax(%arg0: tensor<2x16x32xf32>, %dst: tensor<2x16x32xf32>) -> ten
 // CHECK:        %[[D3:.+]] = linalg.generic {indexing_maps = [#[[$MAP]], #[[$MAP1]]], iterator_types = ["parallel",
 // CHECK-SAME:     "parallel", "reduction"]} ins(%[[ARG0]] : tensor<2x16x32xf32>) outs(%[[D2]] : tensor<2x16xf32>) {
 // CHECK:        ^bb0(%[[IN:.+]]: f32, %[[OUT:.+]]: f32):
-// CHECK:          %[[D8:.+]] = arith.maximumf %[[IN]], %[[OUT]] : f32
+// CHECK:          %[[D8:.+]] = arith.maxnumf %[[IN]], %[[OUT]] : f32
 // CHECK:          linalg.yield %[[D8]] : f32
 // CHECK:        } -> tensor<2x16xf32>
 // CHECK:        %[[D4:.+]] = linalg.generic {indexing_maps = [#[[$MAP]], #[[$MAP1]], #[[$MAP]]], iterator_types =