The builder was taking a carefully ordered list of movement group nodes and
their suggested changes, and squashing the changes all together in a single
dictionary for .edit(**kw).

Now we calculate the edit_order to take the movement_group order into account.

New simulation movements don't have a .divergence_scope attribute, so
movement groups should not assume their presence when checking scope to
adopt a divergence.

Instead of expanding one below it that might not have been modified at all.

cause running out of memory so enforce some restrictions by default which will prevent this.

This updates several alarms after commit 9c94ae13
("Remove default limit on builders' ZSQL select methods").

If there are more simulation movements, the rest is processed next building time.

Proper way is to call a global builder with a 'limit' parameter.

Use of catalog to get related simulation movements from delivery lines/cells is
unreliable.

Until now, for any new code written for simulation, we often had to be careful
not to call getDeliveryRelatedValueList too early, usually by deferring code
to activities with complicated dependencies. Race conditions are difficult to
avoid by developping this way, because for a given delivery, there are
potentially so many events that happen at the same time, involving:
- simulation, amongst causality, expand, building, solving (including split)
- alarms, user actions, external interfaces, chains of activities
- several related deliveries and simulation trees

This commit enables ZODB-indexing of related documents for 'delivery' base
category, making getDeliveryRelatedValueList safe and fixing unlink of
deleted delivery lines/cells.

Existing activity dependencies are left unchanged because builders only uses
catalog and local building needs to find all simulation movements.

This must prevent the useless download of bt5 which are already installed
And so this is supposed to reduced the number of bt5 installed by configurator and make it faster

- duplicated default categories were dropped when the list of categories was
  passed without their base categories
- accept any iterable for 'category_list'

Also simplify sample HTML in test_safeHTML_conversion

this leads to a difference between generated cache id for set and get for *same* function call
thus eliminating image caching.
Examples of generated cache ids:
(get) -> 39600:5:formatpngquality45.0
(set) -> 39600:5:displayNoneformatpngquality45.0

This is a fix for 841b0206 which started to
return "" for empty cells, whereas it's supposed to return None.
This also fixes the 20 minutes slowdown introduced when parsing a category
spreadsheet, such as in the first page of configurator.

This reverts commit 991a6467.
Module is already allowed without this change.

The same mistake seem to be present in other parts of this test.