Add "type" identifier to error rate serialization

[thequilo]

Adds a key "type" to serialized error rates. This fixes ambiguous cases where saved results from DI-cpWER and ORC-WER could not be discriminated, and makes identification easier from the files alone.

In [1]: from meeteval.wer.wer.orc import OrcErrorRate

In [2]: from meeteval.wer.wer.error_rate import ErrorRate

In [3]: er = OrcErrorRate.zero()

In [4]: serialized = er.asdict()

In [5]: serialized
Out[5]:
{'error_rate': 0,
 'errors': 0,
 'length': 0,
 'insertions': 0,
 'deletions': 0,
 'substitutions': 0,
 'reference_self_overlap': None,
 'hypothesis_self_overlap': None,
 'assignment': (),
 'type': 'orc-error-rate'}

In [6]: ErrorRate.from_dict(serialized)
Out[6]: OrcErrorRate(error_rate=0, errors=0, length=0, insertions=0, deletions=0, substitutions=0, assignment=())

The old heuristics for detecting the error rate type are kept for backwards compatibility.

Also fixes a few minor bugs.