More flexible HDF5 support (includes h5py)

[d-chambers]

In my experience (although I haven't tested it lately), pytables is faster than h5py for common read/write operations, and our current indexing scheme makes heavy use of pytables database-like features. However, pytables doesn't support all of the hdf5 datatypes we encounter in DAS files. For example, when reading a terra15 file spews about 8 warnings like this:

DataTypeWarning: Unsupported type for attribute 'primary_dataset' in node 'velocity'. Offending HDF5 class: 8

This is because pytables doesn't support enumerations. These warnings are annoying, but as long as the user doesn't need the value of one of these attributes it can be worked around. However, as I was trying to add support for the Brady Hot springs data (see #278), I found those HDF5 files use linked coordinates, which pytables does not support. As a result, these files simply cannot be read, or at least some important coordinates cannot be accessed.

To remedy both of these issues I propose the following:

1. We take h5py as an official dependency. It appears it only requires numpy, so it should only add a single dependency.

2. We create an abstraction in dascore/utils/hdf5 which unifies data and attribute access for both h5py and pytables (so that either one can be used as a "backend" but the class behaves the same regardless). We could create two called HDF5Reader and HDF5Writer, and these would behave the same as pytables or h5py open files.

3. We modify the stream/buffer reuse stuff mentioned here to return instances of HDF5Reader or HDF5Writer rather than a pytables.File. We could then make it so the reader can be specified as a parameter. EG

# specify read method which uses pytables backend
def read(self, resource: HDF5Reader['pytables`]):
    ...


# specify read method which uses h5py backend
def read(self, resource: HDF5Reader['h5py`]):
    ...

4. We modify the terra15 FiberIO to use h5py to get rid of those pesky warnings
5. We implement the Brady Hot Springs parser.

[d-chambers]

closed by #310