better tests

once we finish cleaning up the code (if it's not already sufficiently clean), we need better tests to verify things are working-- i.e. for some benchmark data, we should verify that performance is at least up to some minimum threshold (that way performance will never gets worse without us realizing).

for example:
- need a way of checking whether things are "working" (i.e. on some sample dataset, can we reconstruct well?)
- check each "condition" (brain object locations are a subset of model locations, brain and model locations match, brain and model locations partially overlap, brain and model locations are distinct)
- also include some checks for when things are expected to break (e.g. not enough locations)
- our checks should go beyond merely checking data types-- they should check that we get the "right answers" up to some threshold similarity or agreement