Type page token/example list truncates multi-word sentences after highlighted span

[fcbond]

On e.g. http://127.0.0.1:5004/type/aj_-_i_le with erg_singlish selected, an
example sentence shows as "Are you kiasi or ?" — it should be
"Are you kiasi or what?". The trailing word(s) after the highlighted
span/token are being dropped in the token list view.

Need to find the rendering path for example-sentence tokens on the type page
(likely in web/db.py's sentence/gold retrieval or the JS that highlights the
matched span) and fix the off-by-one/truncation that loses trailing tokens.

[fcbond]

Root cause found and fixed in ffd7a7a: `gold2db.py`'s `extract_span` only
read the FROM/TO of a terminal's first token when computing the
character span to slice out of the original sentence text. For a
multiword lexical entry spanning more than one input token — like
Singlish "or what" (lexid `or_what_adv1`, tokens "or" + "what") — this
truncated the stored `sent.word` to just the first token's substring
("or"), even though pydelphin's own `terminal.form` already had the
correct joined form ("or what").

Fixed to take the min FROM / max TO across all of the terminal's
tokens. Verified by reprocessing the erg-singlish gold treebank into a
scratch database: sentence 24 ("Are you kiasi or what?") now stores
wid=3 as `'or what'` instead of `'or'`.

This needs a rebuild to take effect on the live site — unlike #59/#60,
this data is baked into the `sent`/`gold` tables at build time by
`grm2db.py`/`gold2db.py`, not computed on the fly. The next
`compile.sh` run (or a targeted rebuild of erg-singlish and any other
grammars with multiword lexical entries in their treebank) will pick it up.
Leaving this issue open until that rebuild happens.

[fcbond]

Root cause fixed in gold2db.py's extract_span (commit ffd7a7a, etc/ltdb repo): it now takes the min FROM / max TO across all of a terminal's tokens instead of just the first, so multiword lexical entries like "or what" get their full span. Covered by 7 new tests in tests/test_gold2db.py.

Verified the fix is correct in isolation, but checked the live build/DBS/ERG_2025.db just now and the sent.word column for or_what_adv1/or_what_tag still shows just "or" — the database predates the fix and needs a rebuild (re-running gold2db.py's process_tsdb over the gold treebank profiles, e.g. via grm2db.py) before this is actually resolved for users. Leaving open until that rebuild happens.