Factor out mutation-setting logic from 'Connection'.

[tseaver]

Toward #514.

[coveralls]

Coverage remained the same when pulling 5f28d27 on tseaver:disuse_connection_save_delete_from_batch into 7c01fcc on GoogleCloudPlatform:master.

[dhermes]

@tseaver

1. This is just a part of the third bullet in Decouple Transaction from Connection #514? I.E., slowly reduce dependencies on save_entity() and delete_entities()?
2. ISTM delete() and put() are ready to go in datastore.api as you outlined in Push batch / transaction onto a thread-local stack inside 'with'. #518, yes? I can do that if you like or just sit back and watch.

Mostly LGTM other than small comments

[tseaver]

@dhermes: rebased / updated to address your concerns in 8bd5059. PTAL.

[coveralls]

Coverage remained the same when pulling 8bd5059 on tseaver:disuse_connection_save_delete_from_batch into 0525530 on GoogleCloudPlatform:master.

[dhermes]

One concern still dangling at

     if auto_id:
         insert = mutation_pb.insert_auto_id.add()
         auto_id_entities.append(entity)
     else:
         insert = mutation_pb.upsert.add()

Can you add a comment above here explaining that we don't use insert or update
but are aware of their existence and may add in future?

---

Can you address my two questions above? See #527 regarding point 2) above.

[tseaver]

Good catch: 0525530 has the comment, which I hadn't saved to disk :(.

WRT #527: I really don't want to have begin() moved outside the context manager: it removes most of the point of the CM in the first place.

[coveralls]

Coverage remained the same when pulling 6af6e6d on tseaver:disuse_connection_save_delete_from_batch into 0525530 on GoogleCloudPlatform:master.

[dhermes]

LGTM