Delete P/D should not delete contents in other datasets

[wmoore]

If I delete a Project or Dataset and choose to delete contents, the contents should only be deleted if they have no other parents.

Images should remain if they are in any other Datasets and Datasets should remain if they are in any other Projects.

[jmoore]

(In [8264]) Preserving doubly linked Datasets and Images on delete (Fix #3031)

[jmoore]

(In [8265]) Also preserving Plates on Screen delete (See #3031)

[jmoore]

(In [8267]) Failing test for SOFT data loss (See #3031, #3032)

[cblackburn]

(In [8269]) Test added ot check that whether Pixels files are deleted for multiply linked images. See #3031

[jmoore]

(In [8271]) Fixing ordering of no-doubly-linked-content deletes in spec.xml (See #3031)

[wmoore]

I find that if ANY of the images in a dataset I'm deleting in Insight (with delete contents) are also in another dataset, then NONE of the images are deleted (end up with some images as orphans).

[jmoore]

At the moment, all IDs to be deleted are stored by location in the graph:

 /Dataset/DatasetImageLink/Image:1,2,3,6,7
 /Dataset/DatasetImageLink/Image/Pixels:6,7,8
 ...

making it impossible to decipher which sub-elements belong to which element, so that if Image 1 throws a ConstraintViolation (i.e. is doubly linked) but Image 2 is fine, we can't know which Pixels should be deleted.

This will require significant refactoring to the DeleteIds object making it into a tree structure which knows which deletes belong with super-element.

[jmoore]

(In [8292]) Refactoring for re-opened content delete issue (See #3031)

[jmoore]

(In [8293]) Test for re-opened content delete issue (See #3031)

[jmoore]

In r8294:

Extensive refactoring of delete logic (Fix #3031)

The moves the logic of deleting from DeleteSpec and DeleteEntry
to DeleteState (formerly DeleteIds). The requirement was that
for a given row delete, it was clear which higher-level object
asked for that delete so that it could be cleanly rolled back.

[jmoore]

(In [8295]) Fixing DeleteITest compile error (See #3031)

[jmoore]

(In [8296]) Fixing broken FORCE logic (See #3031)

[jmoore]

(In [8337]) Missing test from r8294 (See #3031)

[jmoore]

(In [8356]) Improvement of refactored logic from r8294 (See #3031, #3125, #3130)

EventLogs were getting lost due to rollbacks/releases not being
properly called. This commit introduces a the parent step as a
"finished" marker with a "rollbackOnly" flag set to true when
necessary.