Enforce input validation against schema for patching object attributes

[labo-flg]

Problem

This issue appeared with the new schema definition introduced with #5392.
Patching attributes with editField was previously validated with json schema, which has been removed.

This is not an issue 99% of the time... but we introduced a new use case in #4304 : a non-multiple object attribute.

This kind of object can be updated with either a complete payload, such as

{
  "id": "4b762ee9-b2da-49fe-99d7-dfd6b9e1fe33",
  "input": {
    "key": "user_confidence_level",
    "value": {
      "max_confidence": 50,
      "overrides": []
  }
}

... or a partial "patch" update like

{
  "id": "4b762ee9-b2da-49fe-99d7-dfd6b9e1fe33",
  "input": {
    "key": "user_confidence_level",
    "object_path": "/user_confidence_level/max_confidence",
    "value": 50
}

However, nothing prevents the user from sending a complete update with a partial object, such as:

{
  "id": "4b762ee9-b2da-49fe-99d7-dfd6b9e1fe33",
  "input": {
    "key": "user_confidence_level",
    "value": {
      "max_confidence": 50,
    }
  }
}

This would still be valid for elastic, and there is no way of preventing this on elastic side (no concept of mandatory field).
The consequences are bad: the object is now partial in DB, but shall be complete for GQL (in my example, overrides cannot be null)

Solution

Enforce a validation against the schema when editing an object field.

This can be done at the earliest time (resolver) or at the bottom of the chain (just before elastic query). TBD.
-> done just before indexing, when the payload is ready.

[labo-flg]

Completed with #6046