Make `chpl_library_finalize()` not exit the Chapel program for CHPL_COMM=gasnet

[bradcray]

This issue is a follow-on to #22739 and #22913. Specifically, #22739 requested that chpl_library_finalize() not tear down the program such that a user of a library author could do other things after tearing down the Chapel library. And #22913 refactored the runtime to support this for CHPL_COMM=none (and potentially other comm layers).

However, in reviewing the PR, @ronawho noted that chpl_library_finalize() calls chpl_comm_exit() which calls gasnet_exit() which exits the program. This issue captures the desire to eventually extend the current behavior to have chpl_library_finalize() not exit the program for CHPL_COMM=gasnet.

If gasnet_exit() is not called explicitly, it will be called implicitly at the exit of a GASNet program, so my proposed path forward here would be to have chpl_comm_exit() not call gasnet_exit() explicitly, and to rely on this implicit behavior to call gasnet_exit() instead. Though we do additional teardown-related actions after chpl_comm_exit(), Elliot seemed to believe that none of them would be problematic if done before gasnet_exit(), suggesting that this should be OK. The GASNet team also seemed to believe that this approach was reasonable.

While here, we might consider renaming chpl_comm_exit() or chpl_foo_exit() more generally to chpl_comm_finalize() to better indicate "wrap things up for exiting" rather than "literally exit".

[bradcray]

Here are some rough notes that led to this issue, from an earlier draft of #22913:

As I see it, we have a few ways we could proceed there:

• We could have the GASNet version of chpl_comm_exit() (which should probably be renamed to chpl_comm_finalize()) be a no-op and rely on the fact that GASNet programs will call gasnet_exit() upon normal program termination. Though this re-orders teardown operations somewhat (in that comm would be destroyed after the mem and topo layers rather than before), Elliot seems to think that, for GASNet, this reordering wouldn't matter anyway and that it's more important for cases like ugni.
• Nearly equivalently (but more explicitly), we could split the chpl_comm_exit() call into two calls: chpl_comm_finzalize() and chpl_comm_exit() and only call the latter if we're truly exiting the program and not finalizing the library.