Consider further tweak to Table._convert_data_to_col

See https://github.com/astropy/astropy/pull/8933/files#r299094991 and the following code snip. Question is whether this might go into `Table._convert_data_to_col` instead of just `add_column`.
```
# If value doesn't have a dtype and won't be added as a mixin then
# convert to a numpy array.
if not hasattr(col, 'dtype') and not self._is_mixin_for_table(col):
col = np.asarray(col)
```
This same preprocessing is not done before the other 2 places this is called (`_init_from_list` and `replace_column`). Not clear right now if that is a mistake or necessary.

Start by reading the referenced PR discussion and the Table._convert_data_to_col entry point. Compare how the shown preprocessing relates to add_column, _init_from_list, and replace_column, and determine whether the differing behavior is intentional. Done means the desired handling across these callers is decided and documented by the appropriate implementation and validation changes.

Written by the indexing model from the issue text.