When we changed V1L to also be able to use other workspaces than the thread
workspace (aws), we (me) overlooked that the rollback in V1L_Close() can only
work with the aws and only makes sense for the aws.

So to support new use cases where V1L outlives the worker thread, we need to
make the rollback optional. As long as the new use case is the exception,
specifically preventing it without changing the rest of the API seems to be the
least intrusive.

Ref 8c872002
Ref febfce2a
Spotted by ws_emu aka workspace SAN https://app.circleci.com/pipelines/github/varnishcache/varnish-cache/6655/workflows/cb155a57-c1f7-4bc1-885c-0a8432613319/jobs/86172