Integrate global tracking accuracy into eval pipeline

[shaikh58]

Currently the eval pipeline outputs an hdf5 file with motmetrics. See models/gtr_runner/on_test_epoch_end() for implementation. Proposing to add new global tracking accuracy metric into this eval file. Will need to update /inference/metrics.py to calculate this metric. Current working implementation is in /vast/mustafa/global_tracking_accuracy.ipynb

[aaprasad]

In order to do this I would just modify metrics.get_pymotmetrics to compute the metric and add it to the metric dataframe that it returns