Optimize performance of LowPassFilter

[mrxz]

While profiling my wasm port wLipSync, I noticed that the low pass filter was the most expensive by a large margin. Luckily, the performance can be improved drastically with some minor changes. After testing if these performance gains are also possible in uLipSync with the Burst compiler, I can confirm that they are! :-)

This PR optimizes the LowPassFilter function by swapping the inner and outer loop and eliminating the if statement. The following table shows the performance improvement when measuring the entire algorithm (rough measurements):

	ms/it
base	3.22 ms
swapping inner and outer loops	1.48 ms
eliminating if statement	0.82 ms

There is one further optimization possible. Since the next step will be downsampling, the low pass filter can be applied only to the samples that are a multiple of skip in case of DownSampleExact, as the other samples are ignored anyway. But this requires more work to get the skip value used in DownSample before calling LowPassFilter. For reference, here's the commit implementing it: mrxz/wLipSync@5ded0c4

[hecomi]

Thank you for the excellent PR! I'll review it as soon as possible and get it merged.

[hecomi]

Sorry for the very late review, and thank you so much for the excellent pull request!