Add a "nearest non-null" interpolation method for polars.Expr.interpolate

[MarcBresson]

Description

In case we have a series that ends (or starts) with null values, both linear and nearest interpolation will not be able to fill these values.

There could be a new interpolation named "nearest non-null" that would use the nearest non null value. It would give the following examples:

df = pl.DataFrame(
    {
        "a": [4, None, None, None],
    }
)
df = df.with_columns(pl.col("a").interpolate(method="linear"))
df

# shape: (4, 1)
# a
# --
# f64
# 4.0
# null
# null
# null

but

df = pl.DataFrame(
    {
        "a": [4, None, None, None],
    }
)
df = df.with_columns(pl.col("a").interpolate(method="nearest non-null"))
df

# shape: (4, 1)
# a
# --
# i64
# 4
# 4
# 4
# 4

[MarcBresson]

For now, the workaround is to fill null values both backward and forward i.e.

df = pl.DataFrame(
    {
        "a": [None, 4, None, None, None],
    }
)
df = df.with_columns(pl.col("a").interpolate(method="nearest"))
df = df.with_columns(pl.col("a").fill_null(strategy="forward"))
df
# shape: (4, 1)
# a
# --
# i64
# null
# 4
# 4
# 4
# 4

df = df.with_columns(pl.col("a").fill_null(strategy="backward"))
df
# shape: (4, 1)
# a
# --
# i64
# 4
# 4
# 4
# 4
# 4