A few tweaks to the changes #2008 for disposing and normalization.

A few tweaks to the changes #2008 for disposing and normalization.

Alrighty, #2008 did something exceedingly stupid: it lingered *inside the lock*, jamming the connection and backlog during linger. Instead, this moves the thread preservation up in a much cleaner way and doesn't occupy the lock.

Also adds a SpinningDown status so we can see it proper in exceptions, always.

In troubleshooting these 2 tests, I realized what's happening: a really dumb placement mistake in #2008. Now, instead of locking inside the damn lock, it loops outside a bit cleaner and higher up. Performance wins are the same but it's a lot sander and doesn't block both the backlog and the writer for another 5 seconds. Now only the thread lingers and it'll try to get the lock when running another pass, if it gets any in the next 5 seconds.