Fsp hotfixes

[goodboy]

🤦🏼 .. this is what happens when you factor commits out from different layers of a distributed system 😂

Main thing here was,

• use Portal.open_context() to talk to the fsp cluster.. since we ported to that API in #236.
• keep the fsp workers sending on streams since the aggregate throttled graphics update loop won't land until #231.. or thereabouts.

On the flip side, this did find a bunch of stuff to improve / bugfix in tractor so maybe not all bad.

Proposals for tractor from this experience:

• it should be an error to do Context.open_stream() before Context.started()
• there is a spawning case it isn't 'handling where the process handle is not actually ever created and thus teardown logic may run that is unecessary.