an open question should not raise a warning diagnostic

[jarohen]

• An open question is authored on purpose and stays until the question is answered
  allium.openQuestion.present at warning severity reports that intent back as a defect. It's the only analyzer finding that fires on a construct meant to be committed and left in place.

• The cost lands on agents
  One starting cold has to decide per-diagnostic whether it caused it, and "pre-existing, therefore not mine" is the cheap heuristic that works. Training it on a finding that is never actionable is how the real unknownName in the same file gets the same shrug.

• Nothing turns it off
  The severity is a literal at the emission site (analyzer.ts:243), relaxed mode doesn't touch it, and the LSP loads no config — so -- allium-ignore above every open question is the only lever. failOn defaults to warning, so the TypeScript check fails the run on it too.

• Probably wants info, in both checkers
  The Rust CLI (3.5.3) emits nothing for it today, so the severity needs deciding once and applying to both — same shape as allium check: deferred.missingLocationHint warns unconditionally, diverges from TS analyzer (parity doc lists it as Yes | Yes) #20. Keep the code, so open questions stay countable.

• Three language references and the extension's own spec document the current severity
  v1:1001, v2:1170, v3:1434 all say "surfaced by the specification checker as warnings"; allium-extension-behaviour.allium:245 asserts severity: warning.