With complex security scripts to calculate user groups,
calling getPreferredArchive() from DA.__call__ still causes
"maximum recursion depth exceeded" errors when caches are empty.

In order to address the issue, we must review erp5_archive. Instead of that,
this commit first checks whether the archive tool is installed, which is anyway
faster that getting a preference. An expected failing test is also added.

- normalize path to avoid to traverse with wrong value
- review KeyError message when domain is unknown
- use getWebSiteValue when possible

/reviewed-on nexedi/erp5!124

Dependency should be defined directly on bt5 if needed.

'keep_existing' argument of Base._edit has never been used
and we should consider removing it.

When defining a guard on an object, calling
a method such as Base_viewDict on that object
could be denied due to guard acquisition.

- the check is moved to a less performance-critical place
- getPortalObject was forgotten in commit d8e9ed19

- 'apply' causes a break in cProfile results
- 'apply' is deprecated (removed in Python 3)

* reference
* translated_title
* the usual owner_title, creation_date and modification_date
* remove useless listbox_editable_fields that are defined in the field library

This business template does not seem to follow the guideline yet

For example, Python Scripts were:
- dirty in the imported BT (causing all of them to be reinstalled on upgrade)
- compiled to return None

When a currency exchange line is (in)validated, we need to reindex the currency exchange
cells that acquire validation state from the currency exchange line.

This allows defining an exchange rate without end date

https://codemirror.net/demo/activeline.html

since be31425e