FW Lite adapter: resolve lexicon refs for display

First real lexicon integration: implement the port over FW Lite and render glosses that came from a lexicon link.

Scope

The adapter. Implement the port against the Lexicon extension's lexicon.entryService network object, using getEntry / getSense (landed in sillsdev/languageforge-lexbox#2249) and reporting FW Lite's capabilities.

Resolution. Resolve lazily per visible segment, not per project load — a book's worth of refs must not be fetched to render one screen. Cache resolved senses; invalidate on lexicon switch.

The render ladder, which follows the precedence rule already in the model — free-form gloss wins over glossSenseRef / senseRef, and the ref is retained either way:

1. stored free-form gloss, if present
2. otherwise the resolved sense gloss
3. otherwise unglossed

An unresolvable ref is not an error state. Provider offline, entry deleted, and foreign ref (#225) all fall through the same ladder with no error surface and no data loss. The ref stays on the analysis.

Link indicator. Display only — show that a gloss is lexicon-backed, using the third grid row MorphemeBox already reserves for it (src/components/MorphemeBox.tsx). Picking and changing a link is the next item.

Non-goals

No search, no linking, no writes. Read path only.

Note

The deleted-entry and foreign-ref cases overlap #143. Keep the rule in one place — cite it from whichever issue lands second rather than restating it.

Start with the lexicon.entryService network object and its getEntry/getSense operations, then inspect the reserved third grid row in src/components/MorphemeBox.tsx. Trace how visible segments expose gloss and sense references before implementing lazy resolution and caching. Done means the render ladder, lexicon-switch invalidation, fallback behavior, and display-only link indicator work without writes or error surfaces.