fix(jupyter): emit Quarto-canonical cells so captures splice into previews (bd-gthycd33)

[cscheid]

Summary

Fixes bd-gthycd33: executed jupyter output never spliced into the hub-client / q2 preview preview (the {python} cell kept rendering as raw source), while the identical knitr flow worked.

Root cause. The jupyter engine emitted executed cells as bare fences (```python echo + ```{.cell-output}), with no ::: {.cell} wrapper. derive_cell_outputs (capture splice) requires each engine cell to map to a Div.cell in the post-engine markdown, so the cell-output map came out empty and the splice fail-softed to raw source. The Bootstrap CSS (.cell .cell-output-* pre code) never matched jupyter output either.

Fix (emission-side). text_execute.rs now emits the knitr/Q1-parity shape: ::: {.cell} wrapper, {.python .cell-code} echo fence, ::: {.cell-output .cell-output-{stdout,stderr,display,error}} divs (Q1's outputTypeCssClass scheme), and Q1-rule fence sizing (max(3, longest leading backtick run + 1) — fixes latent markdown corruption for backtick-bearing content). Also: array-form mime text (nbformat multiline) is no longer silently dropped, and a cell without a preceding blank line can't glue its div fence onto the paragraph.

Structural anti-regression. The test-only parallel AST emitter (JupyterTransform / outputs_to_blocks) is retired — no production consumer; one emission path remains. Two new fences pin the contract:

• capture_splice_engines.rs — regression pair driving the real engines through record_capture and the exact splice path (jupyter failed with an empty map before this fix; knitr is the control).
• engine_output_parity.rs — 5 cross-engine shape-parity cases (stream, expression value, error under #| error: true, source-only, dependent cells); block-kind + semantic-class tree equality; runtime-gated on both engines being installed.

Verification. Workspace suite 10181 passed; full cargo xtask verify passed; clippy/lint/fmt clean. E2E: q2 render HTML inspected (div.cell, 5 in .cell-output-display); q2 preview in a real browser after the full WASM chain rebuild shows the spliced capture (5 + stream output in the iframe DOM) — the originally-failing surface.

Follow-ups filed (discovered-from bd-gthycd33): bd-ohvl879u (error-policy divergence: jupyter embeds cell errors and continues where knitr fails the render), bd-2me3cslx (pre-existing broken #[ignore]d full-pipeline tests — nested tokio runtime, reproduced at baseline), bd-u996g8g2 (inline {python} expr), bd-5t6wvu7m (real image outputs), bd-vs7sa0qx (execution_count attrs).

Plan with full evidence: claude-notes/plans/2026-07-01-bd-gthycd33-jupyter-cell-wrapper.md.

🤖 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.