ORM was used, resulting in deserialisation/reserialisation of value checked for race
condition during update. In addition to the overhead, this could lead to modification even
if the value didn't change (e.g., because of JSON key order changed).
The raw value is not used to check if the row has not change between `select` and
`update`, making the method more robust.
- `parent_node` and `linked_node` can now be specified in `set_pubsub_node`.
- `get_pubsub_nodes` has been rewritten to recursively traverse a tree and retrieve nodes
matching criteria and their linking nodes/children/items.
rel 463
Add columns to handle `access_model` and `publish_model` and a table for `affiliations`.
With those new data, components handling pubsub can now manage permissions correctly.
rel 462
Use of `pickle` to serialise data was a technical legacy that was causing trouble to store
in database, to update (if a class was serialised, a change could break update), and to
security (pickle can lead to code execution).
This patch remove all use of Pickle in favour in JSON, notably:
- for caching data, a Pydantic model is now used instead
- for SQLAlchemy model, the LegacyPickle is replaced by JSON serialisation
- in XEP-0373 a class `PublicKeyMetadata` was serialised. New method `from_dict` and
`to_dict` method have been implemented to do serialisation.
- new methods to (de)serialise data can now be specified with Identity data types. It is
notably used to (de)serialise `path` of avatars.
A migration script has been created to convert data (for upgrade or downgrade), with
special care for XEP-0373 case. Depending of size of database, this migration script can
be long to run.
rel 443
