Improvements

Added a /v2/mbus/status API endpoint that returns metrics about the state of
the event message bus. This allows for some introspection into what websocket
connections may be doing at that moment.
The UI now has less wasted space when displaying a bunch of cards for data
systems.
Implemented a configurable timeout when making SSH connections to agents. A
lower default now also keeps dead agents from taking up large amounts of
scheduler worker time.
SHIELD now tracks when the last time an agent erred was.
Hidden agents are now sorted under a separate header in the web UI.
Hiding, showing, and deleting agents can now be done from the CLI.

Bug Fixes

We no longer leak the file descriptors and goroutines for detached websocket
clients
Workers can no longer be starved out when sending events to the message bus
if the receiver of the message bus is misbehaving because these event sends
are now asynchronous.
Fixed a bug where a worker could derefence a nil pointer when certain
database selects returned no rows.
The database layer now has more stringent locking, which both avoids certain
threads locking each other out in SQLite, and also makes certain series of
database operations effectively atomic.
A couple of fixups would deadlock themselves out of the database and prevent
fixups from actually running. Now they don't.
Fixups now only run once instead of on every startup, like nature intended.
The agent "Last Checked At" timestamp was being updated when the task was
pulled off the scheduler, whether or not the agent was actually checked
(due to other potential errors).
Named a fixup without a name.
Agents that failed their status checks are now once again marked as such.