Prior to this change when a pipeline processor called another
pipeline, only the stats for the first processor were recorded.
The stats for the subsequent pipelines were ignored. This change
properly accounts for pipelines irregardless if they are the first
or subsequently called pipelines.

This change moves the state of the stats from the IngestService
to the pipeline itself. Cluster updates are safe since the pipelines
map is now a ConcurrentHashMap, and if a cluster update happens
while iterating over stats (now read directly from the pipeline)
a slightly stale view of stats may shown.