fix: Add a fudge factor

[mikebryant]

The reversed code isn't a perfect match for ARM 32-bit floating point assembly. This should allow us to cope with things that are a little out of range due to floating point weirdness

I've checked these issues, and this allows them to compute a pattern:

Fixes #74

Fixes #96

Fixes #63

Fixes #60

Fixes #40

[ZZZZzzzzac]

isn't 5 too large? if it's float point issue there should be only 1 or 2 missing
or let user input fudge factor, 5 makes it's hard to rule out some subpattern

[mikebryant]

Tried 1 at first - when I was going through the issues of people reporting stuff it needed more to cover all the cases.

I suspect the issue is more with the rates, where multiple small floating point errors accumulate to make a larger difference in the final number