Use ONT style for reading/writing fast5 data

[kdoroschak]

Implement raw data writing using a similar strategy to fast5_research. We are already doing parts of this, but updating all of it to be as similar to ONT as possible is a good idea.

• For reading strings: use np.char.decode()
• For writing strings: my_string.encode()
• For writing raw data: verify how they handle dtypes, especially for compound datasets