[MLIR] `promote-buffers-to-stack` fails to promote small allocation

`promote-buffers-to-stack` fails to convert a small heap allocation (`1xi8`, 1 byte) into a stack allocation (`memref.alloca`) even though `max-alloc-size-in-bytes` is set to a sufficiently large threshold (512 MB). The expected optimization (replacing `memref.alloc` with `memref.alloca`) is missed.

**Input Program**

**input.mlir**

```llvm
module {
func.func @threshold_overflow(%value: i8) -> i8 {
%c0 = arith.constant 0 : index
%buffer = memref.alloc() : memref<1xi8>
memref.store %value, %buffer[%c0] : memref<1xi8>
%result = memref.load %buffer[%c0] : memref<1xi8>
return %result : i8
}
}

```

**Command**

```bash
mlir-opt input.mlir -promote-buffers-to-stack="max-alloc-size-in-bytes=536870912"
```

Version: 6f54fe6b6fb22f8a335fd19b94049fc769dfea25

Start by locating the MLIR implementation and tests for the promote-buffers-to-stack pass, then reproduce the issue with the provided input.mlir and mlir-opt command. Done means the 1xi8 memref.alloc is converted to memref.alloca when max-alloc-size-in-bytes is set to 536870912, with a regression test covering the case.

Written by the indexing model from the issue text.