[spanner] Use explicit transactions in db-tests

[pjenvey]

We still have duplicate sync versions of the SpannerDb::commit/rollback/get_transaction methods methods vs their _async (e.g. commit_async) counterparts. This seems due to the sql_request request method's usage, which can implicitly begin a transaction, yet would be inconvenient if every invocation required an .await.

The implicit beginning of a transaction is only used by db-tests ~~(delete_all also uses it! #615)~~, so we should be able to solve this in a different way (maybe the db::tests::support::db function could call begin for the tests? at least under Spanner)

[pjenvey]

The db-tests could call begin themselves, but it probably make sense to provide the test w/ a test_transaction method that does this and rolls everything back at the end, similarl to the http_transaction method now used by the handlers.

[fzzzy]

Yes, we absolutely should do this, it will make the code so much simpler and the tests will test things that are closer to reality.

[pjenvey]

~~More than just the tests use implicit (sync) transactions, all the /info/* endpoints do too because they lack a collection. See #768~~ (#768 now fixed)

[pjenvey]

Another avenue to investigate w/ this issue: would this remove the need for RUST_TEST_THREADS=1 under diesel?

It's not clear why this is required for testing w/ diesel (I know the diesel code base uses it and they've recommended it). I have a theory: begin_test_transaction doesn't rollback. Diesel does have a test_transaction method that takes a closure that does a rollback, however -- would switching to it (or doing the equivalent ourselves) negate the need for RUST_TEST_THREADS?

We might want to ping diesel's gitter about this.