Limitations on using coupe in subplot

Currently, one can only use **coupe** in a subplot if one of these conditions are true:

1. If the **coupe** command is the first in a panel then it needs to set **-R** explicitly and not via **-A...+r**
2. Another plot command with **-R** is given as the first panel command.

This makes it not possible to show a bunch of crossections using the automatic region calculations. Some possible actions we can take:

1. Document this limitation in **coupe** and possibly **subplot** man pages.
2. Isolate the region-calculation in coupe into a separate _gmt_coupe_region_ function that could be used to determine w/e/s/n when subplot requires it.

I think a _gmt_coupe_region_ is the only hope here since the standard way we determine region from data will fail since it will just return the lon/lat extent and not distance/depth...