Worker thread pool and command line changes

[ghjm]

• Change plugin worker interface to use a thread pool and thread-safe queue
  • This isolates plugins from the asyncio event loop, and makes them simpler.
  • This breaks all existing plugins.
• Change recv() to return the whole message, not just the raw_payload
  • This also breaks all existing plugins.
• Add an EOF message type that indicates all response data has been sent
• Create a category of 'ephemeral' commands that will exit when EOF is received
• Use /tmp/receptor instead of /var/lib/receptor for ephemeral commands
• Connect to localhost:8888 by default for ping, status and send
• Change default ping behavior to count of 4 with 1-second intervals

[matburt]

We need to update the merge target for this PR to devel from master

[ghjm]

Added more stuff to this PR:

• Updated file buffers so manifests are kept in the node directory; otherwise, tmpdir cleanup doesn't work. (And this seems to be the intent - I think the current behavior is just a mistake.)
• I was using an incorrect method to mark receptor instances as ephemeral; I have fixed this.
• Ephemeral instances now get removed from routing, rather than going to cost=100, at shutdown.
• The output of receptor status, while still human readable, is also now valid YAML so it can be machine parsed if desired.
• Security fix to avoid receptor nodes being able to call arbitrary methods in worker modules. You now have to explicitly mark a method as receptor exportable.

[adamruzicka]

When I make a silly mistake in the plugin code (for example type asnycio instead of asyncio) the execution silently dies when hitting it and the error does not get propagated to the console.

Logs: https://p.teknik.io/qgHQD

[ghjm]

The exception issue is fixed @adamruzicka.