This fixes number of Jupyter bugs.

Tests passing here: 

https://nexedi.erp5.net/test_result_module/20161117-147BEEA9

/reviewed-on !198

Same as for commit c3450f14:
BTrees don't accept None as key anymore.

Traceback (most recent call last):
  File "Products/ERP5/tests/testAuthenticationPolicy.py", line 538, in test_05_HttpRequest
    self.assertFalse(person.isLoginBlocked())
  File "Products/ERP5/mixin/login_account_provider.py", line 68, in isLoginBlocked
    return method(**kw)
  File "ZODB/Connection.py", line 857, in setstate
    raise ConnectionStateError(msg)
ConnectionStateError: Shouldn't load state for 0x6c90 when the connection is closed

Try to import module before it is added to environment this way if user tries to import non existent module Exception is immediately raised and doesn't block next Jupyter cell execution.

Adds the support of filtering with Ledgers for the generation of the French Fiscal Report (aka FEC).

/reviewed-on nexedi/erp5!195

If user using Jupyter with ERP5 kernel is simply passing a comment inside a cell a server side error is raised thus this commit properly fixes it.
Clean up & style.

In case uid=0, its already being treated as int, so there is no
need to check for type and add time to performance.

__getitem__ function for LazyIndexationParameterList eats up a lot of
performance. This patch tries to simplify the code to improve
performance.

Fix coding style and typos.

Apart from some code cleanup, it can override catalog_index.

Only display desktop mode on big screens.

When we have a transaction with both sections from the same group, we
only check accounting periods for the sides which actually define
accounting. When there was a source or destination organiation defined
at the delivery level it was acquired on the lines and in that check we
wrongly considered this to be defining accounting.

The recently added --frontend_url parameter breaks all existing runUnitTest
scripts that don't accept it, and because this option is irrelevant for many
of them, we don't want to force everyone to update their test suites.

Similarly, parallelizing tests is not worth the effort for some test suites,
and currently, their runUnitTest scripts contain ugly lines like:

    parser.add_argument('--node_quantity', help='ignored', type=int)

These 2 parameters are now passed to runUnitTest only if it supports them,
like it is already the case for --firefox_bin and --xvfb_bin.

Otherwise, the user sees nothing at all whenever she's not allowed to view
at least 1 test suite.

This fixes commit 63b3ccda

ConnectionError does not have status_dict

revert a change submitted by mistake