Fix "memory leak" by closing file objects

[groutr]

The SingleHDF5ToZarr class will consume unsightly amounts of memory over extended periods of time. I tracked this memory usage down to the fact that the class does not close the hdf5 file. Adding a SingleHDF5ToZarr.close() method that closed h5py file and the associated fsspec file object solved the slow leak.

This cut memory usage from >1GB to less than 300MB after processing a large number of hdf5 files. The close method can be invoked directly or via a contextlib.closing context.

The example gen_json function would like the following with these changes to ensure resources are properly cleaned up.

from contextlib import closing

def gen_json(file_url):
    with fs.open(file_url, **so) as infile:
        with closing(SingleHDF5ToZarr(infile, file_url, inline_threshold=300)) as h5chunks:
            # inline threshold adjusts the Size below which binary blocks are included directly in the output
            # a higher inline threshold can result in a larger json file but faster loading time
            variable = file_url.split('/')[-1].split('.')[0]
            month = file_url.split('/')[2]
            outf = f'{month}_{variable}.json' #file name to save json to
            with fs2.open(outf, 'wb') as f:
                f.write(ujson.dumps(h5chunks.translate()).encode());

[martindurant]

Can you please add something to the docsting saying that the user should call .close() to free up memory? Should a default __del__ implementation call it too?