Would it be possible to cast an expression into a specific datatype without casting the source columns?

Consider the following frame:

df = pl.DataFrame({"A": [1, 2, 3], "B": [4, 5, 6]}, schema={'A': pl.UInt8, 'B': pl.UInt8})

┌─────┬─────┐
│ A   ┆ B   │
│ --- ┆ --- │
│ u8  ┆ u8  │
╞═════╪═════╡
│ 1   ┆ 4   │
│ 2   ┆ 5   │
│ 3   ┆ 6   │
└─────┴─────┘

Now using an expression:

df = df.with_columns((pl.col("A") - pl.col("B")).alias("C"))

┌─────┬─────┬─────┐
│ A   ┆ B   ┆ C   │
│ --- ┆ --- ┆ --- │
│ u8  ┆ u8  ┆ u8  │
╞═════╪═════╪═════╡
│ 1   ┆ 4   ┆ 253 │
│ 2   ┆ 5   ┆ 253 │
│ 3   ┆ 6   ┆ 253 │
└─────┴─────┴─────┘

This is obviously correct in comp-sci terms but not practical. Can I only resolve this by first casting “A” and “B” to a signed integer or is there a better/more concise way?