Don't close the whole mpsc channel when one sender is closed

[Nemo157]

I've confirmed locally that this fixes #1440, all senders are given a chance to send before the channel is closed.

There's probably a nicer way to do this with less unwrapping, suggestions welcome otherwise I might take another look whether I can clean this up a bit tomorrow.

[carllerche]

Could the change include tests?

[ebkalderon]

Looks pretty good. Should this PR also include similar changes to UnboundedSender as well? I haven't tested myself to see if it fails in a similar fashion to #1440.

[Nemo157]

It does, both have a disconnect method added and both Sink implementations are updated to use it. I'll add test cases covering both as well.

I have thought of a cleaner way to do this, move all the internal implementation details into an InnerSender struct (although maybe a different name to avoid having .inner.inner fields) which Sender and UnboundedSender contain in an Option, that way disconnecting a specific sender is just dropping that inner struct, and there should be no unwrapping in the internal functions.

[ebkalderon]

@Nemo157 Just wondering, is this PR still being worked on? It appears to be finished.

[Nemo157]

It's all done from my side (but with the All Hands happening this coming week, there's a good chance there's no one available to review for a bit).

[cramertj]

Thanks for the fix!