TFRecords improvement

[RukuangHuang]

Closes #230.

Several improvements related to the use of TFRecords.

• added overwrite argument to indicate whether to overwrite the tfrecord files when tfrecord config is not the same. This solves issue TFRecord dataset not being updated #181 .
• added method to save tfrecord files and function to load tfrecord files as a TF Dataset object, which could be passed to model.fit or model.predict direclty. This solves issue Save/load TFRecord Datasets #230 .
• This PR also fixes some minor bugs.

To save a TFRecord dataset:

from osl_dynamics.data import Data

data = Data(...)
data.save_tfrecord_dataset("path_to_directory")

Note if overwrite=False and a previous tfrecord_config.pkl exists, the files won't be rewritten.

To load a TFRecord dataset:

from osl_dynamics.data import load_tfrecord_dataset

dataset = load_tfrecord_dataset("path_to_directory", batch_size=batch_size)

# Then it can be passed directly to train a model
model.fit(dataset)

[cgohil8]

This looks great!