Correctly handle display usage in cells

[MichaelHatherly]

No description provided.

[MichaelHatherly]

From #44:

I was copying quarto tests to use our native runner, and this failed:

---
title: Julia Subfig Test
engine: julia
---

```{julia}
Pkg.activate(temp = true)
Pkg.add(name = "Plots", version = "1.40.1")
```

## Julia Crossref Figure

```{julia}
#| label: fig-plots
#| fig-cap: "Plots"
#| fig-subcap:
#|   - "Plot 1"
#|   - "Plot 2"
#| layout-ncol: 2

using Plots
plot([1,23,2,4]) |> display

plot([8,65,23,90]) |> display
```

See @fig-plots for examples. In particular, @fig-plots-2.

So they are using display twice to get a multi-column figure output. As we don't hook into the display system but only render return values, this does not work. The question is if we want to make it work this way, or offer an alternative.

[MichaelHatherly]

The question is if we want to make it work this way, or offer an alternative.

Probably just hooking into display, unless you can think of a nicer approach.

[jkrumbiegel]

can we currently add arbitrarily many outputs of a given MIME type to the ipynb we output?

Probably just hooking into display

Yeah either that or with a special type MultiOutput or so that would need to be made available. I guess easier to do the display route

[MichaelHatherly]

Yeah either that or with a special type MultiOutput or so that would need to be made available.

Yes, that's an option, though if a user evals a single cell into a REPL they'll not have that type defined, which would probably be annoying.

Just following what IJulia does is probably the most straightforward thing to do currently.