[Firestore] Refactor Python Transaction layer to better match other SDKs.

[crwilcox]

The Transaction API is vastly different: Instead of offering a Transaction type and a run_transaction() function, a transaction object has to be passed as an optional argument to all WriteBatch methods.

b/76423649

[tseaver]

run_in_transaction is a Spanner-only construct. datastore offers Transaction as a subclass of Batch, and handles the get-a-transaction-id dance in its __enter__. See: https://googleapis.github.io/google-cloud-python/latest/datastore/transactions.html

[crwilcox]

Assigning myself. I need to figure out what this exactly means.

[tseaver]

FTR:

• Spanner uses a "unit of work" callback, and then runs it in a loop under the covers to handle the various ABORTED bits which can be triggered during either reads (by the UoW callable) or at commit time.
• Datastore exposes using its Batch and Transaction objects as context managers: commit happens if the with block exits without raising an exception.

Firestore basically follows Datastore's pattern.

[crwilcox]

I think matching the datastore semantics makes the most sense here.

[crwilcox]

@tseaver, if you can prioritize looking into this? Other clients have get and getAll

https://github.com/googleapis/nodejs-firestore/blob/d307e465aaf8a2709ae0822075488ecb481babf7/dev/src/transaction.ts#L184

So, we would like to add a transaction.getAll(documentReferences) call. Also probably a transaction.get(documentReference) call.

node also supports queryReferences to get. It may be good to do the same in python.