if it's useful to anyone else here's the method we came up for this import:

added a new column to the source spreadsheet (A) for id nos

created a mapping to import these objects with relevant metadata (ignoring representations for now)
created an index spreadsheet (B) to the directories of images we've placed in the import directory. Added a reference table in this spreadsheet (B) to match the id no from spreadsheet (A) with folder names. can go into the method we used for automating this if anyone needs it. ended up with something that looks like:

4A. created a new mapping solely for elements ca_objects.idno and ca_object_representations, both with the option "skipIfEmpty": 1

4B. the ca_object_representations mapping uses refinery objectRepresentationSplitter with parameters:

{ "objectRepresentationType": "front", "attributes": { "media": "³", "idno": "³" }, "dontCreate": "1" }

4C. existingRecordPolicy is set to merge_on_idno

Imported spreadsheet B using this second mapping and all images were found and associated with the previously imported records by idno.

we tried specifing mediaPrefix for each subdirectory of images but never got it working. images only were correctly imported after we removed the mediaPrefix parameter completely, looks like subdirectories in the import folder are scanned for matching files automatically...maybe the import runs faster if the prefix is set?

in reference to the second question above about setting the preferred label: the images imported with the label automatically defaulting to the filename minus the extension, looks like that's just the default behavior...