Set the max-age value to 4h instead of 10min for the one-day-max policy
(ie, 24h/6, like one-hour-max uses 10min max age).

The idea is to reduce backend access of nearly static web sites,
while still allowing changes without waiting for too long before it is propagated.

If user group are recalculated before changing the workflow state, it will only return the same security group
See nexedi/erp5@c00c3636

`createSession` method from the OAuth2 Authorisation Server Connector needs
to access client value. Fetching this value from the session is not needed
since it is already stored in a local variable.

Malevolent users may decide to only - and repeatedly - present an otherwise
valid refresh token, causing the issuance of a new access tokens everytime,
likely along with new refresh tokens, causing many ZODB writes.
Avoid this by pushing the token expiration date by one lifespan accuracy,
so there can only be one write per session per lifespan accuracy period.

This partially reverts 8a336dc5 (erp5_accounting: Allow
Assignor manage Accounting Periods, 2024-09-16) for the restart
transition, it is intentional that only Assignor can restart
an accounting period that have been closed.
The idea was to support a scenario where re-opening a period
that was closed can not be done directly by the Assignee but
needs validation from the assignor.

The check was made on the blob response type, which is set from the
Content-Type header returned by the server, but Safari has a different
interpretation of the charset parameter from the mime type, with a
content type set to application/json;charset=utf-8 like Base_redirect
does today, safari creates a blob with type application/json;charset=utf-8
and this was not detected as redirection and the json returned by
Base_redirect was downloaded. Fix this by checking only the essence
of the type.

This also revealed a potential problem when actually downloading json
files, in that case we also check that we have the X-Location header,
that is supposed to be set by Base_redirect before interpreting the json
and when it's not present we force download.

Follow up of ff624fd2 (ERP5Workflow: newly added permission should be
acquired for all existing states., 2024-11-04) and cbef6282 (ERP5Workflow:
make sure not create duplicate permissions, 2024-11-05)

Fix a problem introduced in ff624fd2 (ERP5Workflow: newly added
permission should be acquired for all existing states., 2024-11-04),
visible in a test failure

before, all permissions became acquired in all states once we update permission list.

See merge request nexedi/erp5!1998

 Don't add to total_contribution_relief if employer price is not set (None).

It was not running because we had two methods with same name

This reverts commit 5e21f77f.

This was done too quickly based on a wrong assumption that simulation
movements to build would always be in planned state and that we could
have an efficient way of selecting them by catalog with index on
portal_type and simulation state, but it does not work this way.

Maybe the change is useful for something else, but since we don't have
any use case for now, let's just revert.

See merge request nexedi/erp5!2001

See merge request nexedi/erp5!2003

In 1b555dbf (py2/py3: Make Products code compatible with both python2
and python3 (nexedi/erp5!1751)., 2024-07-16) we changed the code in a
way that on python2 it would first node text to a python2 str and only
later convert it to unicode, but setting text to a str during the first
step can fail with:

    ValueError: All strings must be XML compatible: Unicode or ASCII, no NULL bytes or control characters

This change to set directly unicode on py2.

and move the implementation of SimulationMovement.getSimulationState
here. This makes it possible for rules to generate simulation movements
that are not necessary planned.

Applied rules must have a rule as specialise

`sender@customer.com <sender@customer.com>` used in the test is not a
valid email address. We have updated to python3.9.20 which comes with a
fix for CVE-2023-27043 and no longer allow this kind of broken addresses.

Replace the address with a similar valid address,
`"sender@customer.com" <sender@customer.com>`, that was probably the
original intention of this test.

On python3, the type of selected columns depend on the data type from
mariadb side, VARCHAR will be str, BINARY/BLOB will be bytes, etc

These SQL method select path that is first evaluated from a variable
that is NULL and in that case, mariadb seems to select LONGBLOB as data
type:

    MariaDB [test]> set @defined_as_null=null; drop table if exists tmp; create table tmp as (select @defined_as_null); show create table tmp;
    +-------+------------------------------------------------------------------------------------------------------------------------------------+
    | Table | Create Table                                                                                                                       |
    +-------+------------------------------------------------------------------------------------------------------------------------------------+
    | tmp   | CREATE TABLE `tmp` (
      `@defined_as_null` longblob DEFAULT NULL
    ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_general_ci |
    +-------+------------------------------------------------------------------------------------------------------------------------------------+

By casting to CHAR in SQL, on the python side we always have the str
that we expect here, because this is used as path attribute of a SQL
brain.