Comm reference fixes + Remove __del__ method and add weakref.finalizer

[JDBetteridge]

This was intended to be just checking current status of PyOP2 but I have discovered issues with the way the @atexit finalizer is cleaning up comms.

[JDBetteridge]

I think this is now the output I expect at the end of testing:

STATE0
PYOP2 Communicator reference counts:
| Communicator name                      | Count |
==================================================
| PYOP2_COMM_WORLD_DUP                   |     1 |
| PYOP2_COMM_WORLD_DUP_COMPILATION       |     2 |
| PYOP2_COMM_SELF_DUP                    |     1 |

Freeing PYOP2_COMM_WORLD
Deleting innercomm keyval on PYOP2_COMM_WORLD
Removing compilation comm on inner comm
DECREF PYOP2_COMM_WORLD_DUP_COMPILATION TO 1
Deleting compilationcomm keyval on PYOP2_COMM_WORLD_DUP
STATE1
PYOP2 Communicator reference counts:
| Communicator name                      | Count |
==================================================
| PYOP2_COMM_SELF_DUP                    |     1 |

Freeing PYOP2_COMM_SELF
Deleting innercomm keyval on PYOP2_COMM_SELF
STATE2
PYOP2 Communicator reference counts:
| Communicator name                      | Count |
==================================================

More generally at the end of any script STATE0 can have:

• 1 reference to PYOP2_COMM_WORLD_DUP and up to 2 references to PYOP2_COMM_WORLD_DUP_COMPILATION.
• Likewise, 1 reference to PYOP2_COMM_SELF_DUP and up to 2 references to PYOP2_COMM_SELF_DUP_COMPILATION.

In its current state it looks like we can get rid of the "ignore all potential sources of error if PYOP2_FINALIZED == True", which would be nice!

[connorjward]

I think this is OK but I don't 100% understand what is going on/why this change was necessary.

[connorjward]

Looks fine.