Custom strategy to determine the slice coordinate in the reference space

In #40 a very simple logic of determining the refspace coordinate was implemented (inside of `download_dataset_parallel`). Namely:
1. Let the user choose `detection_xy=(x, y)` which is a fixed point in the image. Be default it is `(0, 0)`.
2. Query the Image-To-Reference API (`xy_to_pir`) to get the corresponding `(p, i, r)` ([link](http://help.brain-map.org/display/api/Image-to-Image+Synchronization#Image-to-ImageSynchronization-Image-To-Reference))
3. The slice coordinate is then just `p` if `axis="coronal"` or `r` if `axis="sagittal"`.

The above is clearly not ideal. See below some other ideas
1. Keep the current algorithm with a dynamic default - finding the middle of the image automatically since it may vary for each section image
2. Query all 4 corners of the image using `xy_to_pir` and let the slice coordinate be their average
3. Use the `section_number` provided by Allen (see #50) together with the slice ?thickness?