fix: guard concurrent accesses to node api

[iand]

Following investigation by @placer14 we suspect the panic in #411 is due to concurrent modifications of the node field in the actor state task. When originally written this code was sequential but recent changes have introduced concurrent accesses to the field, namely when the Close method is called. I speculate this can happen when another task fails while an actor state task is still spawning goroutines. The task failure is detected by the indexer which closes all lenses used by the other running tasks, setting the node field to nil which is then passed to a new goroutine by the still-running task.

This change guards accesses to the node field with mutexes in each of the tasks. The actor state task checks whether the api lens is available before processing and if it isn't returns a non-fatal error for each actor which will be logged in visor_processing_reports for that tipset+task+actor combination.

Also removed the node field from the messages task since it is not used.

Fixes #411

[placer14]

Thanks for putting this together, Ian.