When the variable type is defined by an IN() function, ``add_type_restriction``
removes all types in it that don't match the specified type. Since the list
of types is modified inplace, the iteration must be done on a copy, not on
the list itself.

There was a test for "add_type_restriction + IN()" case but since there
was only 2 types defined in the IN function, the problem was not exposed.

make it look more like our other modules

sphinx 0.6 switched templates to jinja2, which has an incompatible
syntax for macros.  2009, here we come!
Closes #113086.

This is very useful for test

This is much more readable

We need this for set comparison implemented in 2a175b352c78

The old code did:
* forge a string
* eval it as a lamba
* monkey patch classes with the result

We may just drop the logic looping on multiple class by computing class name
dynamically. But this have a different behavior for subclass.

Use set instead of list for domains. This remove some N² complexity. Simplify
the visite_relation code to:

* avoid computation duplication,
* compare string instead of Yams object,
* take advantage of set.

fix Referencable.get_type crash when no solution given and 'is IN(ET1, ET2..) is used. Closes #81865

668:b2e231cbd9da introduces unrelated changes that 1. break tests, 2. make rql>0.30.1 incompatible with cw < 3.13.10