Old: Merged segments are deleted upon successful push, or on IOException.
New: Merged segments are deleted on any Exception, but NOT successful push.

Deleting a merged segment means that the next merge-and-push run will try to
push it again. So we want that to happen if there was any sort of Exception.
We *don't* want it to happen if the merge-and-push was successful, since in
that case, we are just waiting for historical nodes to load the segment. It
might take a while, but there's no reason to re-push while waiting.

separate cleanup of overshadowed vs. unavailable segments

Coordinator drop/remove related stuff.

- DruidCoordinatorCleanup should wait for availableSegments to populate
  before dropping any segments.
- Clarify that millisToWaitBeforeDeleting applies to "removing" rather than
  "dropping" segments.
- LoadRule shouldn't need to wait for the deletionWaitTime before dropping
  excess replicas.

Separate query metrics into scan/time and query/time

Only emit query metrics for uncached queries.

doc for IngestSegmentFirehose

Conflicts:
	docs/content/toc.textile