Split/join tokens that were wrongly tokenized

[imnasnainaec]

If we tokenize two separate words as a single token, the user should be able to split them, and that should be remembered when the book is retokenized in the future.

Dependency update: "remembered when the book is retokenized" is the crux. Token refs are derived from character offsets, so a split/join shifts every subsequent token's ref in the segment and orphans their analysis links. This issue is blocked on the re-anchoring strategy decision (#136); the likely shape here is a persisted per-project list of token operations (split/join at snapshot positions) replayed after tokenization.

Size: L

[imnasnainaec]

Two UI ideas:

• For situations with lots of manual tokenization, we may want a separate mode/view
• For one-off token adjustment, there needs to be ways to do that without confusing with the phrase construction