Typed script-result handles: Results<"agent-output-400">

Followup to tasks/script-result-truncation.md (which renders an inferred TS type for oversized script results). Bonus round: make that type usable in later scripts, not just readable.

Goal:

const data: Results<"agent-output-400"> = JSON.parse(
  await itx.workspace.readFile(".../script-results/agent-output-400.json"),
);

• register each spilled result's inferred type in the typecheck virtual project (apps/os/src/domains/typecheck/virtual-project.ts) under a Results<Id> lookup type, so itx.docs.typecheck validates drill-in scripts against the real shape
• decide lifetime/scope: types for results of the current agent stream only? persisted where? (probably regenerate on demand from the spill file rather than storing type text)
• surface it in the spill notice recipe once it actually typechecks