Update `where` is ignored if followed by `asFrame`

[Jolanrensen]

To reproduce, make a dataframe with a columnGroup containing a nullable String column "A". Then:

df.update { myGroupedColumn }
    .where { it.a == null } // assumingly filtering to only get rows where this holds
    .asFrame {
        update { a }.with { "None" }
    }

Running this, makes all values of "a" in the columnGroup turn to "None" instead of honoring the where statement.

[Jolanrensen]

Could be solved by something like this:

public infix fun <T, C, R> Update<T, DataRow<C>>.asFrame(expression: DataFrameExpression<C, DataFrame<R>>): DataFrame<T> =
    if (df.isEmpty()) df
    else df.replace(columns)
        .with {
            val processedCol = it
                .asColumnGroup()
                .let { expression(it, it) }
                .asColumnGroup(it.name())

            it.updateImpl(df, filter) { row, _, _ ->
                processedCol[row.index] as DataRow<C>?
            }
        }

The updateImpl function was just completely ignored, handling it as just a replace.with without any filter.

I'll make a PR with tests later :)

[zaleslaw]

I will help with the review

[Jolanrensen]

Actually, I see my fix is wrong. Currently processedCol still takes the entire col, which is not expected if you, for instance, want to calculate the mean of some values of the filtered column.
Filter needs to be applied before expression is executed