* Do not add files to Graph.watchFiles after all files have been loaded
  but immediately when starting to load them so that the same watchFile
  logic can be used for successful and unsuccessful builds (and files
  added by plugins are not ignored on errors)
* Do not unwatch files that have been removed if an error occurred but
  only when a build has succeeded
* Do not re-watch files of the previous build when an error occurred as
  those are still being watched; just additionally watch any files
  attached to the error
* Make sure that watching a file automatically adds it to the list of
  "watched" files and only reset this list when clearing unneeded files
* Remove logic to filter for absolute ids for now to restore the old
  behaviour
* Replace some .forEach with for loops for slightly improved performance

given but for any error that occurs during the build phase when there
are already some watchFiles

…all systems