Fixed an edge case with Safari ("infinite" html5 audio -> the end event won't run)

[golddranks]

Fixes #657 . I'm not well-versed with the codebase, so I tried to code only for the bugged case (checks to call _ended() only if the audio duration was indeed said to be infinite), but it might be, of course, that there is more related, unaccounted edge cases.

[golddranks]

In particular, I'm not sure about removing the event listener after the event is run. I did it because the other event listeners did so too. Are the _nodes re-used (replayed)? After the end is reached, Safari updates the _node.duration to reflect the real duration, but of course this isn't reflected in _parent._duration, so that might cause problems. A review needed!

Btw. to reproducing Safari setting the duration to infinite isn't too easy: it happens to me only if I shut down the server, let the loading fail, and then boot it up, and try loading the same sound again. Safari then sends a Range: bytes=0-1 HTTP header, if to which the server doesn't respond correctly (or doesn't support Range requests), Safari sets the audio duration to infinite.

[golddranks]

Okay, I did one change: it now checks if the duration was set to infinite, and updates the parent _duration field and the default sprite, so the subsequent calls to play() will end as normally.

[golddranks]

Rebased the commits on master. @goldfire, do you think you could pull these in? They fix an edge case on Safari.

[BuddhaBing]

This is still an issue on Safari. Could this fix be merged in please?

[golddranks]

I'd love to see this as merged too. I wonder if this project is abandoned / not maintained?

[goldfire]

@golddranks The project is definitely maintained and not abandoned. We've just gone through some major crunch with a big game launch on Steam last week and I've barely had time to think about anything else. I'll be catching up on howler issues/pr's in the coming weeks.