Compiler plugin sometimes renders `ValueColumn` as frame column

In operations like `update` `.withNull()`

a `FrameColumn` can turn into a `ValueColumn?>` (because nullable framecolumns are not allowed). This is not reflected in the schema produced by the compiler plugin, even though the column is correctly recognized as a nullable dataframe-value column:

A similar problem can occur for `ValueColumn>`, such as with `convert` `asColumn`, `DataColumn.createValueColumn(dataframes)`, or [`toDataFrame { preserve(DataFrame::class) }`](https://github.com/Kotlin/dataframe/issues/1852).

(Though, I might like the idea of blocking any creation of `ValueColumn>`, pushing users to `FrameColumn`)

There may be more cases where this happens, let's find and gather them here

Start by reproducing the schema mismatch in the operations named here: update withNull(), convert asColumn, DataColumn.createValueColumn(dataframes), and toDataFrame { preserve(DataFrame::class) }. Trace the compiler plugin's handling of FrameColumn and ValueColumn> and gather other cases. Done means the affected schemas accurately reflect the resulting nullable dataframe-value columns, with regression coverage for the reproduced cases.

Written by the indexing model from the issue text.