[python] Handling of column_names is inconsistent in ExperimentAxisQuery.to_anndata

[bkmartinjr]

In most of the SOMA API, column_names handling is consistent -- the returned result for a DataFrame read will include the union of:

• columns requested via column_names
• plus, any columns referenced in the value_filter string
  This is true for DataFrame.read(), ExperimentAxisQuery.obs(), and ExperimentAxisQuery.var().

There is a bug in ExperimentAxisQuery.to_anndata(), which does not include the columns referenced in the value_filter. It should work the same as the other interfaces.

The bug is in ExperimentAxisQuery._read_axis_dataframe(), which is overly aggressive in throwing away columns.

Example:

In [15]: next(mouse.obs.read(column_names=['soma_joinid'], value_filter='tissue_general=="lung"'))
Out[15]: 
pyarrow.Table
soma_joinid: int64
tissue_general: large_string
----
soma_joinid: [[1919575,1919576,1919577,1919578,1919579,...,2658971,2658972,2658973,2658974,2658975]]
tissue_general: [["lung","lung","lung","lung","lung",...,"lung","lung","lung","lung","lung"]]

In [16]: with mouse.axis_query("RNA", obs_query=soma.AxisQuery(value_filter='tissue_general == "lung"')) as query:
    ...:     print(next(query.obs(column_names=['soma_joinid'])).to_pandas())
    ...: 
        soma_joinid tissue_general
0           1919575           lung
1           1919576           lung
2           1919577           lung
3           1919578           lung
4           1919579           lung
...             ...            ...
127305      2658971           lung
127306      2658972           lung
127307      2658973           lung
127308      2658974           lung
127309      2658975           lung

[127310 rows x 2 columns]

In [17]: with mouse.axis_query("RNA", obs_query=soma.AxisQuery(value_filter='tissue_general == "lung"')) as query:
    ...:     print(query.to_anndata("raw", column_names={'obs': ['soma_joinid']}))
    ...: 
AnnData object with n_obs × n_vars = 127310 × 52392
    obs: 'soma_joinid'
    var: 'soma_joinid', 'feature_id', 'feature_name', 'feature_length'