Python 3 fixes: Round 3

This PR fixes a few more issues with running iDQ with python 3.

• OVL had a few stray dict.has_key(key) which was changed to key in dict.
• Reading channels from veto lists had a few issues since they're encoded as byte arrays rather than strings when reading back from the numpy arrays. All the relevant channel calls now have explicit channel.decode('utf-8') to handle this.
• A few more casts of hdf5 attribute strings to np.string_(str).
• plots.py: iter.next() -> next(iter).
• reports.py: Empty quivers have their start and end times explicitly set so that calling append_quiver() on globbed quivers doesn't cause errors due to comparisons with None.
• Change how seeking is done in DiskReporter._tail to avoid io.UnsupportedOperation. See https://stackoverflow.com/a/51131242.
• KafkaConsumer.offset_for_times() expects an int for the offset, so there's now a cast to ensure this.

I've tested a few combinations, batch/stream, OVL/sklearn, python2/python3, with/without timeseries and with/without reports.