fusion tests from test_opt

[Qazalin]

test_llama remains in external_test_opt because model.forward calls realize internally.

[github-actions]

This branch currently is behind tinygrad/master. The line count difference bot is disabled.

[Qazalin]

The models tests take the longest:

1.33s call     test/test_schedule.py::TestSchedule::test_enet
1.25s call     test/test_schedule.py::TestSchedule::test_enet_se
1.17s call     test/test_schedule.py::TestSchedule::test_convnext
0.97s call     test/test_schedule.py::TestSchedule::test_resnet
0.64s call     test/test_schedule.py::TestSchedule::test_vit

I think we can keep them in external_test_opt.py and run once in GPU=1.

This picks the the more specific fusion tests from external_test_opt.

[Qazalin]

in test_schedule "fold" is "fuse" for consistency.