Incorrect type when dividing literal series of UInt64 with Int64.

[ritchie46]

Polars version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of Polars.

Issue description

Floor division results to a float type, where Uint was expected.

Reproducible example

s = pl.Series(values=[1], dtype=pl.UInt64) 
pl.select(pl.lit(s) // 2)

shape: (1, 1)
┌─────┐
│ │
│ --- │
│ f64 │
╞═════╡
│ 0.0 │
└─────┘

Expected behavior

shape: (1, 1)
┌─────┐
│     │
│ --- │
│ u64 │
╞═════╡
│ 0 │
└─────┘

[ghuls]

It is not limited to just //:

In [11]: pl.select(pl.lit(s) + 0)
Out[11]: 
shape: (1, 1)
┌─────┐
│     │
│ --- │
│ f64 │
╞═════╡
│ 1.0 │
└─────┘

In [12]: pl.select(pl.lit(s) * 0)
Out[12]: 
shape: (1, 1)
┌─────┐
│     │
│ --- │
│ f64 │
╞═════╡
│ 0.0 │
└─────┘

In [13]: pl.select(pl.lit(s) - 0)
Out[13]: 
shape: (1, 1)
┌─────┐
│     │
│ --- │
│ f64 │
╞═════╡
│ 1.0 │
└─────┘

[ghuls]

Somehow the python integer is not cast to same dtype than the series, resulting in a pl.UInt64 and pl.Int64 -> pl.Float64 conversion as which needs to happen when you would try to get the supertype of a pl.UInt64 and pl.Int64 series:

In [21]: s1 = pl.Series(values=[1], dtype=pl.UInt64)

In [22]: s2 = pl.Series(values=[1], dtype=pl.Int64)

In [23]: s2 + 1
Out[23]: 
shape: (1,)
Series: '' [i64]
[
	2
]

In [24]: s1 + 1
Out[24]: 
shape: (1,)
Series: '' [u64]
[
	2
]

In [25]: s1 + s2
Out[25]: 
shape: (1,)
Series: '' [f64]
[
	2.0
]

It works when split in 2 chained statements:

In [27]: pl.select(pl.lit(s).alias("a")).select(pl.col("a")  + 0)
Out[27]: 
shape: (1, 1)
┌─────┐
│ a   │
│ --- │
│ u64 │
╞═════╡
│ 1   │
└─────┘

[ritchie46]

Yes, I think I need to check the type_coercion optimizer to see if the literal fits into the columns dtype.