feat(render): add --strict to promote warning diagnostics to errors

[cscheid]

Closes #220.

Summary

Adds a --strict flag to q2 render that treats warnings as errors, for CI use. Warning diagnostics are promoted to error severity (kind: Warning → kind: Error) on the completed ProjectRenderSummary — the single seam every structured diagnostic flows through before anything is printed or counted — so the text output, --json-errors wire output, the error/warning counts clause, and the exit code all agree. Because promotion happens at the aggregation boundary rather than at emission sites, any warning added anywhere in the pipeline in the future (stages, transforms, Lua filters, project post-render) automatically participates in strict mode with no per-call-site opt-in.

Design notes, alternatives considered, and the architecture map are in claude-notes/plans/2026-07-02-strict-mode-warnings-as-errors.md (braid strand bd-yjs54ptg).

Semantics

• Without --strict: behavior unchanged.
• With --strict: everything still renders (the flag changes the outcome, not the control flow), then diagnostics print with error severity and the command exits 1 if any former warning or error is present. Info/Note are untouched.
• Orthogonal to --fail-fast, which keeps meaning "stop at first failure" — a promoted warning does not stop the render.
• CLI-only policy: q2 preview / hub-client stay deliberately lenient per Decision D1 (bd-creo).

Implementation

• ProjectRenderSummary::promote_warnings_to_errors() in quarto-core covers all four diagnostic sources (pass-1/pass-2 failure diagnostics, project diagnostics, per-output diagnostics via a new OutputDiagnostics::diagnostics_mut).
• should_exit_nonzero gains a strict arm; it defensively also fails on surviving warnings so the gate is correct independent of promotion order.
• The externalized quarto-error-reporting crate is untouched — severity policy stays consumer-side.

Testing

• TDD: 7 CLI integration tests (crates/quarto/tests/integration/strict_mode.rs) written first and confirmed failing, spawning the real q2 binary against an unresolved-crossref fixture; plus 7 unit tests for the promotion method and the exit gate.

• cargo nextest run --workspace: 9884 passed. Full cargo xtask verify (incl. WASM/hub-client leg) green.

• End-to-end, output inspected:

  $ q2 render warn.qmd --strict     # warn.qmd references @fig-nonexistent
  Error: unresolved crossref `@fig-nonexistent`: no target with this identifier was found.
  1 error
  # exit 1; warn.html still written
  
  $ q2 render warn.qmd --strict --json-errors
  {"$schema":".../json-diagnostic.json","kind":"error","title":"unresolved crossref ..."}
  

Follow-up

bd-zcjtaz78 (discovered during this work): Error-kind diagnostics on successful outputs print as error but exit 0 without --strict — the non-strict gate deliberately keeps that behavior in this PR.

🤖 Generated with Claude Code

[posit-snyk-bot]

✅ Snyk checks have passed. No issues have been found so far.

Status	Scan Engine	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues
✅	Licenses	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.