More generic NamedTuple handling

[haberdashPI]

As noted in #41 by @ararslan:

• It should be possible to support arbitrary NamedTuples that have :start and :stop at minimum, possibly as part of a larger set of keys, e.g. (; start=Second(1), stop=Second(5), event="explosion"). AFAICT the only real tradeoff is that you get an ErrorException rather than a MethodError if you're missing :start or :stop.
• IMO, the :start and :stop keys needn't be Nanosecond-valued or even Period-valued; in the same manner that e.g. TimeSpan(0, 10) works, we could convert to Nanosecond as needed by defining

  start(nt::NamedTuple) = Nanosecond(nt.start)
  stop(nt::NamedTuple) = Nanosecond(nt.stop)

  which should provide the same set of guarantees.

• ...actually, we could even define an untyped fallback method that does exactly the above, which would cover NamedTuples for free. Hmmm.

[kleinschmidt]

I think automatically supporting anything that can be converted to Nanosecond is a huge footgun. It's not hard to imagine someone creating a timespan with e.g. Ints giving start/stop in seconds or milliseconds without fully understanding what's going on internally and having mysterious, silent breakage (when comparing to other spans with nanosecond-based times).

[kleinschmidt]

(of course, the same goes for the TimeSpan(0, 10) constructor)

[haberdashPI]

Oh good point... mm... for now maybe the #41 should only accept tuples with Period values.

[ararslan]

Yeah, that's a very good point. Scratch that part then.