prf:algorithm: support continuous line numbering for algorithm2e-style algorithms (nested blocks restart at 1)

[mmcky]

Summary

algorithm2e (LaTeX) numbers every algorithm line continuously (1…N) and conveys nesting — while / for / repeat bodies — through indentation. When such an algorithm is rendered in MyST (e.g. inside a {prf:algorithm} directive), the natural representation is an ordered list with a nested ordered list for each loop body. But CommonMark restarts numbering at 1 for every nested list, so the loop body is numbered 1, 2, 3… again instead of continuing the outer count.

Example

1. $t \leftarrow 0$
2. input $X_0$
3. **while** $t < \infty$ **do**
   1. observe $X_t$
   2. choose action $A_t$
   3. receive reward $r(X_t, A_t)$
   4. draw $X_{t+1}$ from $P(X_t, A_t, \cdot)$
   5. $t \leftarrow t + 1$
4. **end**

Renders as: 1, 2, 3, then 1, 2, 3, 4, 5 (restarted), then 4.

algorithm2e (PDF) renders as: continuous 1 … 9, with the loop body indented but its line numbers continuing the sequence.

Request

An enhancement so algorithm-style content can render with continuous line numbering across nested blocks while keeping the indentation that shows loop/conditional structure — i.e. algorithm2e parity. This may take the form of a {prf:algorithm} option, a list-numbering mode, or directive-level support; the exact mechanism is open.

Context

Surfaced converting the QuantEcon Dynamic Programming Volume I book; reported downstream at QuantEcon/book-dp-public#31 (item 1). The converter side is tracked at QuantEcon/claude-latex-to-myst#169 — this upstream issue is the likely home if continuous-numbering-with-indentation cannot be produced in plain MyST/CommonMark.

[mmcky]

Per the routing in #88, the fix is presentation-side and now tracked at QuantEcon/quantecon-theme.mystmd#119 — CSS counters keyed off a :class: on the directive (the option already exists via commonDirectiveOptions), with the converter side at QuantEcon/claude-latex-to-myst#169. This issue stays open as the cross-repo tracker; no engine change is planned unless myst-driven PDF export of these books becomes a requirement.

[mmcky]

Routed out of the engine by #88 — retained here only as a cross-reference tracker.

Noting it on the issue itself, since the routing decision was recorded in #88 and this one still reads as engine work.

The AST already represents an algorithm faithfully as nested ordered lists, so continuous line numbering across nesting is presentation rather than representation, and no engine change is needed. The opt-in hook also already exists — prf:algorithm accepts :class: through commonDirectiveOptions.

Work now lives at QuantEcon/quantecon-theme.mystmd#119 as CSS counters: one counter scoped to the algorithm root, incremented on every descendant li, with no reset on nested ol, so nesting keeps its indentation while the line count continues.

One question there is still open and decides whether any converter change is needed at all — house style for every prf:algorithm versus an opt-in class. I recommended house style on corpus evidence: 24 algorithms across the three books, 21 nested and 3 flat, where continuous numbering renders identically for the flat ones, so the rule covers 100% of the corpus with 21 improved and 0 regressions. Downstream converter tracker: QuantEcon/claude-latex-to-myst#169.

Known limit recorded in the routing: a CSS-only approach covers HTML output; a myst-driven PDF export would need engine-level representation, which no current book needs. If that changes, this issue is the place it comes back.