fail on loading cp3 model in cp4

[mrariden]

Check the state_dict after loading to verify that it's not a CP3 model. This will need to be revised if we ever implement concurrent support for CP3 and CP4

Passing tests on:

• GH partial tests
• Ubuntu full tests
• mac partial tests

[mrariden]

resolves #1300

[mrariden]

Now the transformer will check the state_dict during model loading for the W2 parameter that reshapes token into pixel space. If that's not found, a ValueError is raised.

[junliuchengxia]

winddows all fali