[@loopback/sequelize] Duplicate column names in relation queries

Describe the bug

Repository queries with relations that do not explicitly set all the keyTo / keyFrom / through properties will lead to broken SQL queries that contain duplicated title case column names mixed in with the expected camelcase columns.

Example:

SELECT bookid as bookId, reader.id as readerId, ReaderId …

Relates to:
https://github.com/sequelize/sequelize/issues/9328
https://github.com/loopbackio/loopback-next/issues/9591
https://github.com/sourcefuse/loopback4-sequelize/issues/35

Logs

No response

Additional information

Workaround seems to be to go through all the Entity relations and explicitly set all the relation key names but that can be time-consuming and error-prone with a larger project. It seems like it might be possible to mitigate this at the loopback model to Sequelize relation layer.

Reproduction

With these changes, running the tests for the Sequelize extension will replicate the error:
https://github.com/KalleV/loopback-next/commit/3c29d852b46f19f4805d556ae32aee49c79471f1

Turns out it's necessary to define additional "belongsTo" relations in other entities before this happens. The extra relation is added to the "Patient" entity in this case. With this set up, I am seeing the following happen:

• The default relation properties are set by Loopback (i.e. I can see the "keyFrom" is populated as todoListId) but the "keyTo" is undefined:
  
• This leads to undefined being passed as the foreign key to sequelize:
  
• And then that causes Sequelize to assign a Title Case property through it's own default relation column logic leading to a duplicate column name in the database query:
  

Beginne mit den Sequelize-Erweiterungstests und den Reproduktionsänderungen in Commit 3c29d852b46f19f4805d556ae32aee49c79471f1. Untersuche das Einrichten der Patient-Relation sowie die Verarbeitung von keyFrom, keyTo und through, bevor die Abfrage Sequelize erreicht. Erledigt ist die Aufgabe, wenn Relationen, bei denen nicht jede Eigenschaft explizit gesetzt ist, keine doppelten Spalten mehr erzeugen und die Erweiterungstests eine Regressionstestabdeckung enthalten.