feat(dataset-updater): align translation tasks to gpt-5.5 mandate (#192, #299 follow-up)

[jsboige]

What

Enforces the standing gpt-5.5-only translation mandate. The eco-tier gpt-5.4-mini is banned for translations ("plus de fenêtre eco-tier ; pushback si un worker propose 5.4-mini"). This PR flips every gpt-5.4-mini occurrence in DatasetUpdaterRootConfig.cs to gpt-5.5.

Code-only, single file. References the #466 DOC-ONLY investigation docs/investigations/2026-06-14-llm-model-quality-datasetupdater.md, which identified the gpt-5.4-mini residue as drift from the mandate.

Count note: the investigation's pre-edit table stated 10× gpt-5.4-mini. The working-tree file actually carries 12 such occurrences (the doc undercounted — it tabbed the 4 Rules ES/AR/FA/ZH tasks but its "10" summary did not fully reconcile, and Fallacies→EN was still on the eco tier despite the doc's synthesis text). All 12 are translation tasks → all move up, per the "every occurrence" directive.

Exactly which tasks changed (line, name, old → new)

#	Line	Task name	Old → New
1	246	Translate Fallacies to English by branch empty-only 0-shot	gpt-5.4-mini → gpt-5.5
2	309	Translate Fallacies to Russian by chunk empty-only 0-shot	gpt-5.4-mini → gpt-5.5
3	373	Translate Fallacies to Portuguese by chunk empty-only 0-shot	gpt-5.4-mini → gpt-5.5
4	579	Translate Virtues to Russian by chunk empty-only 0-shot	gpt-5.4-mini → gpt-5.5
5	643	Translate Virtues to Portuguese by chunk empty-only 0-shot	gpt-5.4-mini → gpt-5.5
6	692	Generate Virtues Portuguese Wikipedia links empty-only 0-shot	gpt-5.4-mini → gpt-5.5
7	933	Translate Scenarii to Russian by chunk empty-only 0-shot	gpt-5.4-mini → gpt-5.5
8	1001	Translate Scenarii to Portuguese by chunk empty-only 0-shot	gpt-5.4-mini → gpt-5.5
9	1302	Translate Rules to Spanish by chunk 0-shot	gpt-5.4-mini → gpt-5.5
10	1345	Translate Rules to Arabic by chunk 0-shot	gpt-5.4-mini → gpt-5.5
11	1388	Translate Rules to Farsi by chunk 0-shot	gpt-5.4-mini → gpt-5.5
12	1431	Translate Rules to Chinese Simplified by chunk 0-shot	gpt-5.4-mini → gpt-5.5

Count: 12 changed lines (git diff --stat = 12 insertions / 12 deletions; file otherwise byte-identical).

Intentionally left untouched (separate decision for @jsboige)

The 4 gpt-5.4 (non-mini) tasks are FR-native generation / refinement, not translation. The mandate is about translation, so these are left as-is and flagged here as a distinct decision:

Line	Task name	Model (unchanged)	Why kept
61	Update Virtues Taxonomy by chunks 1-shot	gpt-5.4	FR taxonomy creation/refinement
117	Update Fallacies French Description by branch 0-shot	gpt-5.4	FR desc_fr refinement
182	Update Fallacies French example by branch 0-shot	gpt-5.4	FR example_fr creative generation
446	Cleanup Fallacies translations by chunk empty-only 0-shot	gpt-5.4	multi-lang cleanup review (quality tier)

If you want these aligned to gpt-5.5 too (or the cleanup-review one in particular), that's a separate call — happy to follow up.

No runtime / cost effect

All 47 tasks remain Enabled = false. There is no execution or API cost until a deliberate run is enabled.

Validation

• Build + tests: dotnet test Argumentum.AssetConverter.Tests → 159 pass / 0 fail / 5 skip (skips = GSheet/GDrive + Freeplane GUI, expected).
• EOL integrity: git ls-files --eol → i/lf w/crlf preserved (no line-ending flip).
• Diff scope: git diff shows exactly the 12 Model = lines; no other byte changed.

Model distribution after: gpt-5.5 31 → 43, gpt-5.4-mini 12 → 0, gpt-5.4 4 (unchanged), total 47.

🤖 Generated with Claude Code