Make MaxConcurrentReconciles configurable for Agents

[manno]

Currently it's possible to configure the maximum number of reconciles (workers) for some controllers.

It should also be possible to configure it for these controller:

• the agent
• imagescan
• cluster
• clustergroup

What are good defaults? A value of 1 will process events in sequence and avoid conflicts when updating resources. The wrangler-based implementation used 50, however it's unclear if queues behave the same. How do the operators behave with e.g. 2000 goroutines per reconciler? To which degree does it affect memory consumption?

see https://github.com/rancher/fleet/pull/2897/files