Merge in accessor IDs based on search of parent directories, instead of re-creating stack table entirely and reseting index.

Deleting accessors by task ID excludes those that are used later but in a task that is still finished; don't do this in batch runner

Session wasn't updating accessor's info state when it was loaded, and hence not scheduling it for deletion

This reverts commit 3332573d.

Separate out method to read each accessor from file, so that this can be overridden in subclasses if needed