Stop daemons & timers while the operator is on peering freeze

[nolar]

When the operator goes to peering freeze, all the watch streams disconnect from the API and wait until the operator is resumed.

However, this mode was not passed to the daemons & timers subsystem, so they continued running — while not getting any updates on the resource's state from the cluster (because the watch-streams are frozen).

This fix synchronizes the daemons & timers to the global operator freeze:

Once an operator is frozen, all daemons & timers are stopped almost the same way as when the operator exits (e.g. due to SIGTERM/SIGINT) — with all the cancellation backoffs, timeouts, and other procedures of graceful & forced termination.

Once the operator is resumed, nothing is done explicitly: the resumed watch-streams will naturally spawn new daemons/timers for all matching resources to the moment — as if those started to match the filters, all at once. (This will implicitly cover the case of "up to date" relevance: those resources that did not exist before the freeze, will be spawned too; those that has gone, will not be resumed.)

TODOs left:

• Edge case: prematurely stop the stopping of daemons if the operator is resumed before all of them are stopped.
  • Uncertain: just unset the stopping reasons? but what if the daemon's code is already in the existing clause? finish the stopping? then, how is it re-spawned if no new events arrive for the resource after that?
• Tests.

Fixes #673.