[MLIR] `scf-parallel-loop-fusion` misses fusion for identical producer-consumer parallel loops

`scf-parallel-loop-fusion` misses a simple fusion opportunity between two adjacent scf.parallel l oops. The first loop writes vector chunks to `%arg0`:
```llvm
%arg0[i : i + 4] = %arg2
```

The second loop has the same bounds and step, reads the same chunk from `%arg0`, and writes it to `%arg1`:

```llvm
%v = %arg0[i : i + 4]
%arg1[i : i + 4] = %v
```

Since both loops have identical iteration spaces and the second loop directly consumes the data produced by the first loop, they could be fused into a single `scf.parallel`. Currently, the two parallel loops remain separate, leading to a missed optimization.

### Input Program
**input.mlir**

```llvm
module {
func.func @equal_mask(%arg0: memref<16xf32>, %arg1: memref<16xf32>, %arg2: vector<4xf32>) {
%c0 = arith.constant 0 : index
%c4 = arith.constant 4 : index
%c16 = arith.constant 16 : index
%cst = arith.constant 0.000000e+00 : f32
%cst_0 = arith.constant dense : vector<4xi1>
scf.parallel (%arg3) = (%c0) to (%c16) step (%c4) {
vector.transfer_write %arg2, %arg0[%arg3], %cst_0 {in_bounds = [true]} : vector<4xf32>, memref<16xf32>
scf.reduce
}
%cst_1 = arith.constant dense : vector<4xi1>
scf.parallel (%arg3) = (%c0) to (%c16) step (%c4) {
%0 = vector.transfer_read %arg0[%arg3], %cst, %cst_1 {in_bounds = [true]} : memref<16xf32>, vector<4xf32>
vector.transfer_write %0, %arg1[%arg3] {in_bounds = [true]} : vector<4xf32>, memref<16xf32>
scf.reduce
}
return
}
}
```

### Command
```bash
mlir-opt input.mlir --scf-parallel-loop-fusion
```
Version: 6f54fe6b6fb22f8a335fd19b94049fc769dfea25

Start with input.mlir and run mlir-opt input.mlir --scf-parallel-loop-fusion to reproduce the missed optimization. Trace the scf-parallel-loop-fusion pass's handling of adjacent producer-consumer loops and verify that the two matching loops are fused into one scf.parallel while preserving the shown transfers.

Written by the indexing model from the issue text.