core: Fix bug in extractor when having nested variables

[math-fehr]

When having an attribute constraint variable T and a variable U that could infer the
variable T (for instance U = Vector<T>), the constraint U would not realize that it can infer T.

[alexarice]

Seems fine but do you have a use case for this? I didn't include this as couldn't think of an operation that would actually need it.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.27%. Comparing base (2b2f51e) to head (b8da9e1).
Report is 3 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #3850   +/-   ##
=======================================
  Coverage   91.27%   91.27%           
=======================================
  Files         461      461           
  Lines       57656    57678   +22     
  Branches     5571     5571           
=======================================
+ Hits        52623    52645   +22     
  Misses       3609     3609           
  Partials     1424     1424           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[math-fehr]

Seems fine but do you have a use case for this? I didn't include this as couldn't think of an operation that would actually need it.

Yep, I got this in downstream code:

@irdl_op_definition
class StoreOp(IRDLOperation):
    name = "smt.array.store"

    _DOMAIN: ClassVar = VarConstraint("_DOMAIN", AnyAttr())
    _RANGE: ClassVar = VarConstraint("_RANGE", AnyAttr())
    _ARRAY: ClassVar = VarConstraint("_ARRAY", ArrayType.constr(_DOMAIN, _RANGE))

    array: Operand = operand_def(_ARRAY)
    index: Operand = operand_def(_DOMAIN)
    value: Operand = operand_def(_RANGE)

    res: OpResult = result_def(_ARRAY)