bpe-openai: allow lazy loading BPE and feature flag each model

The fully generated `.dict` models take 70mb, which bloats binaries substantially. This is compared to 3.4mb for the targz.

It would be nice to support lazily doing this translation, which takes a modest hit to runtime performance in return for dramatically smaller binary, and the ability to only load certain models reducing memory consumption.

It would also be nice to have feature flags to turn off certain models; this is less important if we add lazy loading, though, as it would just allow you to save 1-2mb off the binary.

The issue names no files, tests, or entry points; begin by locating the bpe-openai generated .dict models and the code that translates or loads them. Done means models can be loaded lazily, individual models can be selected with feature flags, and binary and memory savings are demonstrated without losing supported behavior.

Rédigé par le modèle d'indexation à partir du texte de l'issue.