off-by-ones in mark pos rule deltas

This is evident, for instance, in Kablammo. We expect to sometimes have some off-by-ones in other places, but we don't currently allow any in otl-normalizer output.

- if possible, we should just figure out what we're doing differently in the math
- if that's too annoying, we can fix some off-by-ones in the normalizer output?

```
python3 -m ttx_diff 'https://github.com/Vectro-Type-Foundry/kablammo?cccc120d23#sources/Kablammo.glyphs'
```

Start by running the provided ttx_diff command against the Kablammo source and inspect the reported mark positioning rule deltas. Compare the differing math or normalizer output to determine whether the underlying calculation or only the emitted off-by-one values should change; done means the discrepancy is resolved without allowing unintended off-by-ones in otl-normalizer output.

Written by the indexing model from the issue text.