Close open portals after receiving PortalSuspended

[imalsogreg]

I used some printf debugging to determine that PortalSuspended message is received during a call to wait_until_ready(). This probably means that wait_until_ready() is being called at the wrong time? However, handling PortalSuspended here by sending another Execute command, as the psql spec suggests, solves the multiple-portal issue we see with CockroachDB.

@robo-corg and I are working on a similar patch that puts the handler in a less surprising place. But I'm opening this PR in case you're interested in a quick fix, or in case this helps us see how the protocol handling should be changed.

Fixes #933 and subsumes #944

[mehcode]

I'm fine with either fix going in the next release to make it work for you guys. I think I prefer this one a touch as it doesn't change the public API.

[imalsogreg]

@mehcode Thanks! I think we'll spend a couple more days seeing if we can find solutions that we like a little better. And then we'll check in with you and see what we all think is the right thing to merge.

[imalsogreg]

NB that this patch doesn't work in our bigger codebase. Our testsuite throws

Error: InternalError(encountered unexpected or invalid data: expecting ParseComplete but received CommandComplete)

So we shouldn't merge this PR. In addition to the CI here reporting that it's broken, we see that it's broken in the wild too.
Will continue working on alternatively.