Tasks were previously being put into QUEUED state after they were active
in the ECS executor. This was to store executor state for task adoption
but had the side effect of removing them from the list of running task
instances (which has other knock-on effects). Instead change tasks into
the RUNNING state, and do not remove them from the list of running
tasks.

- DebugExecutor: was overriding the change_state method from the base
executor, but changing no behaviour, so move to using the base executor
implementation
- CeleryExecutor: Plumb through the new param so that the signature
matches the base executor

…state

This allows task adoption to still work while using a more accurate
state for the executor.

There is a task filter earlier in the event processing method that I
missed originally