sql: step the transaction before auto-commit

[ajwerner]

We want the pre-commit validation to see the side-effects of all previous
statements. If we don't step the transaction, we won't see writes which
occurred in the last statement. The shocking thing is that this all worked
before this change. The reason it worked was that the internal executor usage
that happened as a part of logging every schema change resulted in the txn
being stepped. That is a separate, and more complex bug to fix. It will be
described and addressed separately.

Fixes #86132

Release justification: Important bug fix

Release note: None

[cockroach-teamcity]

This change is 

[knz]

LGTM but you need to update the expected output for the enums logic test.

[knz]

needed before #86174 can merge

[ajwerner]

you need to update the expected output for the enums logic test.

Doesn't need a rewrite, was a flake fixed by #86276

[knz]

so this is good to go?

[ajwerner]

Yeah, here goes.

bors r+

[craig]

Build failed (retrying...):

• Bazel Essential CI (Cockroach)

[knz]

I think this is bound to fail due to "TxnCounter":23 in the expected output of TestTelemetryLogging?

[knz]

bors r-

[craig]

Canceled.

[ajwerner]

I think this is bound to fail due to "TxnCounter":23 in the expected output of TestTelemetryLogging?

I dug into this more. It turns out that test is flakey. It does not look at the TxnCounter field, just the SkippedQueries field.

cockroach/pkg/sql/telemetry_logging_test.go

Lines 348 to 350 in fd1b51f

	if expected := fmt.Sprintf(`"SkippedQueries":%d`, expectedSkipped); !strings.Contains(e.Message, expected) {
	t.Errorf("%s: expected %s found:\n%s", tc.name, expected, e.Message)
	}

#86118

bors r+

[craig]

Build succeeded:

• Bazel Essential CI (Cockroach)