Reduce functional test execution time

[justinshenk]

Duration for functional tests can be shortened by using pytest fixtures and reducing epochs from 10 in neural network tests (test_models). Here are some extensive durations:

154.86s call     traja/tests/test_models.py::test_ae_jaguar
85.59s call     traja/tests/test_models.py::test_aevae_jaguar
65.71s call     traja/tests/test_dataset.py::test_time_based_sampling_dataloaders_with_stride_one_do_not_overlap
65.23s call     traja/tests/test_models.py::test_ae_classification_network_converges
65.23s call     traja/tests/test_models.py::test_vae_classification_network_converges
64.47s call     traja/tests/test_models.py::test_vae_regression_network_converges
64.38s call     traja/tests/test_models.py::test_aevae_regression_network_converges
64.34s call     traja/tests/test_models.py::test_ae_regression_network_converges
42.49s call     traja/tests/test_dataset.py::test_time_based_weighted_sampling_dataloaders_do_not_overlap
39.99s call     traja/tests/test_dataset.py::test_id_wise_sampling_with_short_sequences_does_not_divide_by_zero
39.11s call     traja/tests/test_dataset.py::test_id_wise_sampling_does_not_put_id_in_multiple_dataloaders
38.83s call     traja/tests/test_dataset.py::test_id_wise_sampling_with_few_ids_does_not_put_id_in_multiple_dataloaders
38.43s call     traja/tests/test_models.py::test_lstm_jaguar
32.18s call     traja/tests/test_dataset.py::test_time_based_sampling_dataloaders_with_short_stride_do_not_overlap
24.18s call     traja/tests/test_dataset.py::test_time_based_sampling_dataloaders_do_not_overlap
21.04s call     traja/tests/test_dataset.py::test_id_wise_weighted_sampling_does_not_put_id_in_multiple_dataloaders
13.25s call     traja/tests/test_dataset.py::test_sequential_data_loader_indices_are_sequential
6.84s call     traja/tests/test_plotting.py::test_plot_prediction