Better cleaning of sessions sub-directories

**Description of the desired feature**

Given recent anecdotal stories on the [forum](https://forum.generic-mapping-tools.org/t/question-about-gmt-session-folder/2436/12) about huge amounts of disk space taken up by long-dead GMT sessions, it seems reasonable to restate some goals and possible solutions to make things better. Some goals:

1. No `gmt_session.######` directory should remain when a session is over
2. No active `gmt_session.######` directory should be accidentally removed

We recently implemented the handling of Ctrl-C interrupts so that the session directory is removed upon exit. However, we have lots of other situations that could need handling:

1. When a GMT module returns with an error code (say, failing to parse a badly formatted **-R** or **-J**)
2. When a non-GMT tool fails in a GMT script

In either case, the GMT script will most likely fail to give a valid result or figure. I see different solutions for these cases:

1. When a GMT module returns a non-zero status we could terminate the session before returning. If the command is in a bash script then the **set -e** part will exit the script so that no other commands will be run. We do this by simply inserting another command in the bailout macros along the lines of `if (code) gmtlib_terminate_session;` Wrappers can also check the error return and take any evasive actions suitable.
2. When a non-GMT command fails we would need the bash script to have set a trap so that an error triggers a `gmt clear this` (or similar command before exiting.

A final _ad-hoc_ cleaning up scheme could be when **gmt end** is called: It could examine the contents in the sessions directory, get a list of all the PPID embedded, and then determine if these are current processes or not. If not a current process then we delete the corresponding directory.

Happy to entertain comments on these possibilities!