fix read loop with shutdown

[dpatti]

When you shutdown the connection and we close the reader, a surprising
thing happens under certain conditions: the read loop wakes up, asks for
the next read operation, and is told to yield. It tries to yield, and
then the state machine raises -- you shouldn't be yielding if the
connection is closed!

The reason for this is our explicit return of [`Yield] in
_final_read_operation_for. As mentioned in my lengthy comment, I think
the real issue here is that if we want to circumvent Reader.next, we
have to make sure we're not yielding when the reader cannot take any
more input anyway.

I thought of fixing this in a few ways, like adding explicit yielding
and continuing mechanisms to the reader at request boundaries, which is
something that has come up in the past as being a bit tricky to reason
about. I also considered just tossing this check at the top of
next_read_operation, but it feels like the kind of thing that I'd
later ask "why is this necessary?" I want so badly for the reading and
writing behaviors to be parallel, so I did something that is meant to
feel less permanent.

This test also fails before the refactor-request-queue branch. I found
it curious, but then realized that Server_connection.shutdown is not
used in either of our runtimes, so it's not too surprising that we
didn't have decent coverage of it.

Note on testing

One of the solutions I tried was to change the peristent_connection
branch to also check closed state. But then I realized that you could
have a closed reader but still want to process more connections, so you
wouldn't want to simply shut the whole thing down. I added a missing
test that will fail if you do the wrong thing, now.

[dpatti]

The test came from investigation in #209