limit pending requests

[bwidtmann]

during our long term playback tests we realized a lot of playback crashes. we realized that it always happens when there are a lot of rejected requests in parallel. therefore we limited the amount of pending requests in PlaybackTimeRule.

Additionally we made retry attempts and interval configurable. If you set the attempts to a very high number, the player now - with limiting pending requests - can resume playback correctly after lost internet connection (fixes Issue: #599)

[dsparacio]

@bwidtmann , just an FYI, I have a side branch I am working on some concepts that remove the entire rejected request idea. We found that rejected request create unreliable buffer growth readings to be able to do accurate heuristics based on buffer occupancy vs throughput. This is because the buffer will "burst" when seg X is rejected, segment X +1 +2 get appended in front of segment X but do not add length to the buffer level until seg X (that was rejected) gets re-downloaded and appended. Now the buffer level "burst" three segment lengths at once instead of just the expected one segment length. By removing the rejected queue we have perfectly accurate growth. We have a working concept but many things need to be tested before we commit. Most likely nothing will be committed for 1.5 instead work this idea in during a refactor 1.6 release.

[bwidtmann]

@AkamaiDASH as you mentioned this is for 1.6, I think we should provide a quick fix for now. This fix is not that big but increases long term streaming stability a lot. So I think we can do this in parallel without getting a lot of merge conflicts in 1.6

[dsparacio]

@bwidtmann Totally agree. I was going to say we should fix just this as an issue for 1.5 but you already did that. So thank you! 👍

[KozhinM]

@bwidtmann, the fix looks fine, thanks!