Firestore: Add 'should_terminate' predicate for clean BiDi shutdown.

[tseaver]

Stream exceptions which pass this predicate will cause clean shutdown of the BiDi thread.

Closes #7826.

Note to reviewers: I recommend looking at this commit-by-commit.

[tseaver]

I can also add should_terminate support for Pubsub, if desired (and someone can confirm the set of gRPC status codes which should cause clean shutdown).

[crwilcox]

Is there a reason should_recover couldn't be used for this? Semantically, should not recover and should terminate seem the same? Do we need another argument.

Also, note, immediately following this merge, we should ship api_core and rev min versions for firestore and pubsub

[tseaver]

@crwilcox See #7826. The goal is to avoid logging an error when the stream is shut down due to "expected" events., typically due to a grpc.StatusCode.CANCELLED error response

• The existing should_recover returns True if the error is "recoverable", i.e., the stream should save any pending send and reopen itself.
• The new should_terminate returns True if the error should cause the stream to shut down cleanly, without reopening.

[plamut]

The code changes look good to me.

I was able to reproduce the issue as described, but after re-installing api_core and the firestore changes from this PR, I observed a different error:

(venv-3.6) peter@black-box:~/workspace/google-cloud-python/firestore (pr_temp)$ python reproduce-7826.py 
on_snapshot()
Thread-ConsumeBidirectionalStream caught unexpected exception 'NoneType' object has no attribute 'target_change' and will exit.
Traceback (most recent call last):
  File "/home/peter/workspace/google-cloud-python/firestore/venv-3.6/lib/python3.6/site-packages/google/api_core/bidi.py", line 653, in _thread_main
    self._on_response(response)
  File "/home/peter/workspace/google-cloud-python/firestore/venv-3.6/lib/python3.6/site-packages/google/cloud/firestore_v1/watch.py", line 429, in on_snapshot
    target_change = proto.target_change
AttributeError: 'NoneType' object has no attribute 'target_change'

It might be caused by my local/project setup, but it might as well be unrelated to the error reported in the ticket (which seems to be fixed by this PR).

[tseaver]

@plamut Maybe the patch from this PR allows a final None to be passed through on a cancelled stream, where previoiusly it would have errored out with the traceback from #7826?

[tseaver]

In any case, if it is possible that on_snapshot can be called with None as the response, we need to handle that (by an immediate return, I think).

[plamut]

Applying the changes from #8687, the AttributeError is now gone, and the fix for the original issue still works.