Revert peakfinder changes.

@luke.davis tested Py2 and the tdv-use-original-resampler branch and found only small differences in the number of background triggers being recorded in bankstats.

He traced the difference back to a '<' -> '<=' change I'd made in Peakfinder. That change handles the case where a large number of snrs have the exact same value, e.g. if the entire stream is 'infinity', which I encountered in a particularly buggy version, and possibly with engineering run data!

I think it's a useful refactor and a good fix, but it should be a separate item for review. We want to be able to demonstrate stable, long-term, bit-for-bit identical results for the Py3 upgrade.