* HistoryManager (especially FileHistory) can now cache its own instance
  of ItemManager which is the main optimization here. This gets rid of
  listing all metadata .txt files from the history collector directory
  and and reading them all in on every history sampling.
  * Caches are invalidated upon any sign of changes to the set of
    subscribed items. FileHistory uses Java NIO WatchService to track
    any changes to .txt files in the history collector directory and
    invalidates the item/item manager caches upon such an event.
  * Remove the need to always reconstruct ItemManager if the set of
    items does not change - which it doesn't during normal simulation.
  * Allow internal caching of items for HistoryManager.getItems()
    which also doesn't change unless the set of subscribed items
    changes.
* ItemManager support indexing the contained items by a single field
  name. This allows faster search lookups of items based on the
  `variableId` field in history sampling.

gitlab #1092