Merge method for SFE

[lambdamoses]

Kind of like cbind but not quite. In cbind, for all methods that I know of, the number of rows must match. For merge, the rows don't have to perfectly match. The user interface should follow that of base R's merge for data frames whenever applicable, giving options for full join (I think should be default, unlike in base R), left join, right join, and inner join. Give an error when the row names of the two SFE objects have no overlap, and warning when the overlap is small. rowGeometries are subsetted or padded with empty geometries accordingly. rowGeometries of the same name but different content in the two SFE objects are treated the same way as in the SFE cbind method. colGeometries and annotGeometries are simply concatenated as in cbind.

[lambdamoses]

I feel like because the merge method is not specific to the spatial aspect of the data, this is more appropriate as a PR to batchlor (which already has intersectRows, effectively inner_join) or tidySCE to merge SCE objects that don't have the same genes.

[lambdamoses]

Generally full join filling with 0 is a bad idea, so this is low priority. The correct way is to fill with NA, but this has downstream implications. There already is SummarizedExperiment::combineCols to do a full join. But still cbind and combineCols can get fragile for SFE when there are many colGeometries, since it won't work when some colGeometries have different non-spatial attribute columns. However, I don't expect this scenario to happen very often since I don't often use columns of colGeometries. In any case, it's better to cbind before further analyses. Even without the spatial info, cbinding or merging have implications to data normalization and interpretation of dimension reduction results.

[alikhuseynov]

I'm just adding here as we discussed.
merge method using inner join, and to deal with mismatches in columns of colGeometry, colData, reducedDims, localResults would via SummarizedExperiment::combineCols.