Adjust algorithm for importing modules for resolver

[RecurvedBow]

To import a module, it is executed. However, if the module contains a module which is not imported yet, a ModuleNotFoundError is raised. Adjust the algorithm of the resolver so that missing modules are imported first before importing the actual module.

Due to this issue, the subtyping filter can not be used for unification of trace data of big projects.

[bengsparks]

I would like a minimal reproducible example of this

[bengsparks]

Perhaps this function is what is needed after all.

[RecurvedBow]

Try to do this: Import a module of another project whose module name is in the trace data, can be found in the project (for example: src/folder/a.py -> Import the module with name src.folder.a) and apply the unify subtypes filter. That module imports another module which is not used in the PyTypes project (for example, torch).

The tests of the project to trace itself should be able to be successfully executed without any issues (= trace data can be generated), but the PyTypes project has to import these modules.

A ModuleNotFoundError in _attempt_module_lookup() in resolver.py should occur. If not, mark this issue as obsolete.