fix: remove a stale type: ignore and document value_eq's broad catches

[tschm]

Two quick fixes from the quality assessment in #414, one per commit.

fix(types) — remove a stale suppression (#418)

src/loman/ui/widget.py:352 carried # type: ignore[arg-type] on the **options splat, and ty flagged it as an unused blanket directive — the underlying signature has been tightened since it was added. A stale ignore is worse than none, because it keeps hiding a real arg-type error if one is reintroduced at that call site.

make typecheck reported "Found 1 diagnostic" before this change and "All checks passed!" after.

docs(util) — say why value_eq swallows comparison errors (#420)

Both broad catches in value_eq returned False with no stated reason. The intent is defensible, but it was unwritten — and every other broad catch in the codebase already carries its rationale (computeengine.py:692 logs via LOG.exception; transformer.py:1049 is an optional-extra guard; computation.py:831 has a pragma note; ui/value.py:81 explains that a broken __repr__ must not break the detail panel). These two were the exception.

The rationale comes from the call site rather than from guessing at intent. value_eq has exactly one caller — Computation.insert at computeengine.py:1140 — which uses it only to skip a redundant write on an UPTODATE node. So "not equal" means the value is written and descendents marked STALE: a recomputation that was not needed, rather than a stale graph that looks up to date. That asymmetry is what makes swallowing correct here, and letting the exception escape would instead fail an insert over a comparison the caller never asked to be able to fail.

No behaviour change in either commit — one deleted comment and two added ones.

Verification

Gate	Result
make fmt	PASS — 21/21 hooks
make typecheck	PASS — 0 diagnostics (was 1)
make test	PASS — 1542 passed, coverage 98.94%, unchanged

Closes #418
Closes #420