utils/frame_queue: change pl_source_frame and pl_frame_mix to double

Fixes mixed usage of float/double for timestamp and improves estimate accuracy to avoid drift. There is no reason to use floats in this code.

Improving precision of the estimator seems like a good idea, but I don't see why any of the other floats need to be promoted.

And re: estimator, maybe we can use a smarter summation method?