feat: Adding support for exclude_txn_from_change_streams option

[aakashanandg]

feat: Adding support for exclude_txn_from_change_streams option while creating RW transactions.

This PR introduces support for the exclude_txn_from_change_streams option for RW transactions. This allows developers to selectively prevent specific transactions from being recorded in a database's change stream.

Usage Example:

# This transaction will NOT be captured by any change stream.
Singer.transaction(exclude_txn_from_change_streams: true) do
   # Perform high-volume or low-value writes here.
  Singer.upsert_all(...)
end

Fixes #367

[olavloite]

LGTM, with a few minor nits

[olavloite]

nit: I don't think we need this in the commit request. This only goes in the BeginTransaction request.

[aakashanandg]

I took reference from Ruby client library where this attribute was getting set during the commit call. For ref:https://github.com/googleapis/ruby-spanner/blob/a2e5c8c8b8a909746bbc9aff1b79884ed44db104/google-cloud-spanner/lib/google/cloud/spanner/session.rb#L656

But anyways i will remove this from this commit as it is not required.

[olavloite]

Ah, yeah, I understand that. The difference here is that:

1. A CommitRequest can also start a transaction. This is called a single-use write-only transaction, and in such a case the CommitRequest both starts and commits the transaction. This is sometimes used for bulk writing. The drawback of these transactions is that there is no guarantee that they will be applied only once. It could be that a component somewhere in the network path between the client and the server replays the request, for example due to a temporary network problem leading to an UNAVAILABLE error, and in that case the transaction would be applied twice.
2. Ruby ActiveRecord always starts the transaction first, meaning that the CommitRequest will never start a transaction. So in this case we never need this option here.

[olavloite]

This test does not really appear to test anything relevant. The exclude_txn_from_change_streams flag goes into the BeginTransaction options, but this test sets that flag after the transaction has started, and dos not verify anything regarding the BeginTransaction options.

[aakashanandg]

Got it. At first, I believed this attribute had to be set during the commit call. I've now updated the test to assert the TransactionSelector object assigned during the transaction.begin call instead.

[olavloite]

nit: This SQL statement would in the real world not return 3 rows, it would only return an update count. It would slightly improve the readability of this test to add THEN RETURN * to the SQL string, so it matches what the mock return value is.