Run-order based fallback windows

[jaketmp]

Determine fallback integration windows based on bounds of prior and subsequent samples, rather than global medians

[adwolfer]

The current implementation treats every sample independently (no knowledge of run order) which is a challenge to the definition of FIR as bound by the found peaks of prior and subsequent samples.

It can be envisioned to recalculate FIR as a last processing step once every samples and features have been found