Fix start values of FIIR and Biquad filter to second sample

[theHenks]

In the current implementation of the FIIR and Biquad filters, the first sample is off since the algorithms starts with an initial state of zero(T). This creates unwanted behavior e.g. for the DifferentiatorFilter:

The fix sets the first sample to the value of the second sample to avoid this behavior:

[oschulz]

I don't think this fix will work for all Biquad filters, it works for the differentiation filter because that has a finite-length response. I think we need to initialize s1, s2 and s3 correctly, instead of fixing the first output values after filtering.

[fhagemann]

I don't think this fix will work for all Biquad filters, it works for the differentiation filter because that has a finite-length response. I think we need to initialize s1, s2 and s3 correctly, instead of fixing the first output values after filtering.

@oschulz could you have a look into this? Or recommend how EXACTLY it should be done or how we can be sure it is done the correct way? This filter is kind of critical..

I think the issue here is that the differentiation would usually result in shorter length vectors, but we still force the length of the filtered array to be the same