Option-D w10 follow-ups: scope ANIMAL, ratified terms, protect \ifswedish clamps

[hmiddelk]

Post-merge follow-ups to #948/#949 from BR's review, all verified in a rebuilt compendium-en.pdf.

Changes

1. Drop the inmixning supertyp override — BR fixed the missing comma at source (plan/Plan.scala, 53730d5, closes Concept list: 'inmixning supertyp' fused into one item (missing comma in plan/Plan.scala W10 contents) #952), so the concept-list item is now two concepts (inmixning→mixin, supertyp→supertype) translating on their own. The autotranslate: fix half-translated concept-list terms (topptyp/bottentyp/inmixning supertyp) #949 workaround is removed. (Left the wrong inmixning supterp cache row purged, not restored — it's dead now; say if you'd rather have a literal revert.)

2. Scope the ANIMAL cluster to perFileId["w10-inheritance-exercise"] (was global). Djur also occurs in lect-w11-context's generics demo (class Katt/Hund extends Djur), where a global Djur→Animal produced mixed class Katt extends Animal. Scoping keeps w11 fully Swedish (deferred) while Task 3 still translates. The rest of the cluster occurs only in this file.

3. Mirror leaves code inside \ifswedish clamps untranslated — translateCodeEnvBodies was running renderCodeIds on code inside hand-written \ifswedish…\else…\fi clamps, mutating the Swedish branch into a mixed listing (the hand-clamped Fyle example rendered def makeNoise … print(läte*2)). New Latex.ifswedishRanges (nesting-aware) gives the clamp ranges; both the env pass and inline pass skip matches inside them.

4. Ratified terms (BR): IckeAkademiker→NonAcademic; Doktorand→"PhD student" (title string → codeStr); lowercase akademiker→academic, forskare→researcher (Task 5 REPL val names).

5. Clamp the Fyle räknaLäte solution — the one Fyle block that wasn't \ifswedish-clamped, so väsnas collided (must be makeNoise in Task 3 but holler in Fyle — same file, unresolvable by glossary). Hand-clamped using the Fyle example's established terms (väsnas→holler, läte→call, Fyle→Fowlk) plus new räknaLäte→callCount, antalLäten→numberOfCalls. Re-termed BR's existing Pjodd \else clamp (was callTally/callCount) to the same scheme so both solutions read consistently. Swedish branches kept byte-identical.

Verification

• Full compendium-en.pdf rebuilt: Task 3 (Animal/makeNoise/Mooooo), Task 5 (Person/Academic/Researcher/NonAcademic, "PhD student", val academic/researcher), main Fyle example (Fowlk/holler/call), both Fyle solutions (callCount/numberOfCalls, no mixing), w11 block consistently Swedish, concept list (mixin/supertype separate). No makeNoise+läte mixing anywhere; no callTally left.
• Compile+leak gate: 10 ok, 0 regressions; 293 .tex, 0 code-string leaks.

Notes for BR

• räknaLäte→callCount, antalLäten→numberOfCalls are the two identifiers I coined (BR-ratified) — the rest reuse your Fyle \else terms.
• Pre-existing prose mistranslations in the räknaLäte subtask question ("Before a method…", "a certain frog") are model/cache issues, unrelated to this PR — flag if you'd like them addressed separately.

Phase-1 inline compile gate (#951)

Adds the inline-.tex compile gate BR scoped in #951, as a second check in scratch/verify-mirror-examples.scala:

• Gates the display Scala-code envs the mirror rewrites (Code/CodeSmall/lstlisting) that renderCodeIds actually changes and that aren't inside an \ifswedish clamp (reuses the new Latex.ifswedishRanges, so it gates exactly the set the mirror translates).
• Same regression rule as the autotranslate: fix s-interpolation example-compile break + CI compile gate #950 .scala gate: compile the Swedish body and the rendered English body; regression = Swedish compiles but English doesn't; skip if both fail (neighbour-context/signature-only/script-style bodies self-skip). Strips a leading lstlisting[...] option line first. Uses the mirror's per-file overrides so per-file clusters (ANIMAL) are gated too.
• REPL transcripts deferred to phase 2 (need splitting on scala> prompts); Trace/Output aren't compilable Scala. Both counts are log()-ed so nothing silently reads as "covered".

Tested (not just run clean): 23 self-contained inline bodies compile (non-vacuous); an injected Tomat→Tomato duplicate-class rename is caught (exit 1) while a \ifswedish-clamped copy of the same break is correctly skipped; robust to a missing examples/ dir.

[hmiddelk]

@bjornregnell ready for review.

Seven commits, all verified in a rebuilt compendium-en.pdf (not just the mirrored .tex):

• Scope + bleed fixes: ANIMAL cluster moved to perFileId["w10-inheritance-exercise"] (kills the class Katt extends Animal mix in lect-w11-context); the mirror now leaves code inside \ifswedish clamps alone (Latex.ifswedishRanges), fixing the hand-clamped Fyle example.
• Ratified terms: IckeAkademiker→NonAcademic, Doktorand→"PhD student", lowercase akademiker/forskare→academic/researcher.
• Fyle räknaLäte solution hand-clamped (the one unclamped Fyle block — väsnas collides makeNoise vs holler). Reuses your Fyle \else terms; I coined räknaLäte→callCount, antalLäten→numberOfCalls and re-termed your Pjodd \else (was callTally/callCount) to match — flagging those two as the only new names for your OK.
• Dropped the inmixning supertyp override now that Concept list: 'inmixning supertyp' fused into one item (missing comma in plan/Plan.scala W10 contents) #952's comma is fixed at source.
• Phase-1 inline compile gate (autotranslate: extend the compile gate to inline .tex code (not just \scalainputlisting .scala) #951) as we discussed: gates rewritten Code/CodeSmall/lstlisting bodies (skips \ifswedish, defers REPL to phase 2), autotranslate: fix s-interpolation example-compile break + CI compile gate #950 regression rule. Tested non-vacuous (23 bodies compile) and effective (an injected duplicate-class rename is caught; a clamped copy is skipped). -- . --list prints the per-body OK/skip breakdown.

Clean gate run: example gate 10 ok / 0 regressions; leak-check 293 .tex / 0 leaks; phase-1 23 ok / 30 skipped / 97 REPL deferred / 0 regressions.

Two review notes: (1) the only invented identifiers are callCount/numberOfCalls — everything else reuses existing terms; (2) pre-existing prose mistranslations in the räknaLäte question ("Before a method…", "a certain frog") are model/cache issues unrelated to this PR — happy to address separately if you want.