Ingest fails when features are not 100% the same for both adata objects.

• [X ] I have checked that this issue has not already been reported.
• [X ] I have confirmed this bug exists on the latest version of scanpy.
• [X ] (optional) I have confirmed this bug exists on the master branch of scanpy.

I'm working on comit 63b42e4b (latest master).
I'm not sure if intended or not but it seems like it would be usefull if one were able to ingest data that don't share 100% of all features.

Minimal code sample (that we can copy&paste without having any data)

adata = sc.datasets.paul15()
sc.pp.pca(adata_ref)
sc.pp.neighbors(adata_ref)
sc.tl.leiden(adata_ref)
adata = adata_ref[:, :1000].copy() # assume adata_ref has more than 1000 genes.
sc.tl.ingest(adata, adata_ref, obs='leiden')

Error message

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-37-b3cd11e67810> in <module>
----> 1 sc.tl.ingest(adata, adata_ref, obs='leiden')

~/projects/scanpy/scanpy/tools/_ingest.py in ingest(adata, adata_ref, obs, embedding_method, labeling_method, neighbors_key, inplace, **kwargs)
    125 
    126     ing = Ingest(adata_ref, neighbors_key)
--> 127     ing.fit(adata)
    128 
    129     for method in embedding_method:

~/projects/scanpy/scanpy/tools/_ingest.py in fit(self, adata_new)
    437 
    438         if not ref_var_names.equals(new_var_names):
--> 439             raise ValueError(
    440                 'Variables in the new adata are different '
    441                 'from variables in the reference adata'

ValueError: Variables in the new adata are different from variables in the reference adata

Versions

Start in scanpy/tools/_ingest.py at Ingest.fit, especially the variable-name comparison that raises the reported ValueError. Reproduce the example with a reference AnnData object and a 1,000-gene subset, then inspect the surrounding ingest flow. Done should be defined as ingest accepting the intended non-identical feature sets without breaking the existing matching-feature case.