stream.Run() returns nil on forced shutdown (benthos logs a warning
instead of returning an error). Changed RunAsyncWithErrors to RunAsync
and removed the unused helper method.

Fixes CON-419

StopNow calls StopWithin(0), which returns immediately without waiting
for the stream to stop. The stream goroutine later exits with
context.Canceled when test cleanup cancels t.Context(). The previous
fix (744e120d0) assumed stream.Run() returns nil on forced shutdown,
but it returns context.Canceled. Accept that as a normal shutdown error.

Fixes CON-419

Replace the first 5s fixed sleep with require.Eventually polling for
snapshot output, and reduce the second 5s sleep to 2s. The test was
timing out under the global 5-minute package budget because other slow
tests (1M-doc parallel snapshot, 35s connection timeout test) consumed
the allotted time first.

Fixes CON-420

In update_lookup mode, MongoDB sets fullDocument to null when the
document is deleted before the post-update lookup completes. The code
checked only key presence, not nil value, so a null fullDocument passed
through as a nil message. Now falls back to documentKey when
fullDocument is null for update events, matching the delete-event
fallback pattern.

Also adds a small sleep in the integration test between UpdateOne and
DeleteByID to prevent the race from making the test flaky.

Fixes CON-428

The default 5-minute timeout expired before all 58 tests in the
mongodb/cdc package could complete, causing a package-level panic.

Fixes CON-420

When the benthos read_until idle timeout fires, the framework stops
draining readChan before mongodb_cdc.Close() is called. During this
window, snapshot goroutines block on the readChan send with no reader
and no context cancellation. Add m.shutsig.SoftStopChan() to the
select statements in readSnapshotRange and readFromStream so goroutines
can exit immediately when shutdown begins.

Fixes CON-430

TestIntegrationMongoCDCWithParallelSnapshot was writing 1M documents and
taking ~136s, consuming most of the 5-minute go test timeout. Later tests
like TestIntegrationMongoCDCSchemaValidator had no time budget left and
were killed by the global timeout.

100k documents still exercises 8-worker parallel snapshot adequately
while reducing the test from ~136s to ~18s.

Fixes CON-420

The replica set can take a moment after container readiness before it
accepts client connections through the mapped port, causing intermittent
i/o timeouts. Wrap the initial Ping in EventuallyWithT so transient
failures during replica set initialization don't abort the test.

Fixes CON-452