Add plugin hook for dynamic class definition

[ilevkivskyi]

Fixes #5508

The new hook allows this:

from some_lib import dynamic_base

Base = dynamic_base()

class C(Base):  # No error, the plugin acts and replaces 'Base' with a TypeInfo
    ...

This plugin hook is useful for SQLAlchemy ORM, for user-defined constructs similar to namedtuple(), and maybe we can even re-implement namedtuple() as a plugin at some point.

Here are some comments:

• I don't add basic_new_typeinfo() to semantic analyzer API, because it is hard to find a signature that would satisfy all possible use cases. Instead I add just add_symbol_table_node() and qualified_name(), so that a user can construct TypeInfo manually and add it to the symbol table. We can reconsider this when we will have more experience with this hook, and add a dedicated utility method.
• Semantic analyzer API in semanal_shared.py has add_symbol_table_node(), but I think it is outdated, and should be replaced with add_symbol() which is IMO better (this is unrelated to this PR, but if you agree I will open an issue).
• If you think the added test case is overcomplicated, this is on purpose. The added test case is a PoC implementation of how this would work in SQLAlchemy, I wanted to be sure that new hook will work well also in combination with the other base class related hook (and I would say this may be a typical combination).

[JukkaL]

Looks good, just a few ideas about potential new tests.

[JukkaL]

Also test a few different assignment statements and ensure that the plugin doesn't get run them. Maybe things like A = B = ... and C = <other_function>(). If the plugin doesn't get run, the lvalue shouldn't be a valid base class.

[JukkaL]

Would it make sense to add an AST diff test case for the base class? E.g. if there is no change, the base class shouldn't be triggered.

[ilevkivskyi]

@JukkaL Thanks! I added the tests you suggested.