perf(catalog): trim highlight.js bundle from ~190 to 35 grammars

[fiskus]

Summary

• Replace the convenience highlight.js default import with a shared app/utils/hljs.ts that imports highlight.js/lib/core and registers a bounded set of 35 languages explicitly. Runtime API (hljs.highlight / hljs.getLanguage) is unchanged at every call site, so 7 files only need the import path swapped.
• The registered set equals the keys of Text.js's LANGS map — the only enumerated runtime consumer of hljs in the catalog. python / bash used by Code.tsx callers are already in that set.
• Narrow Code.tsx's hl: string prop to hl: RegisteredLanguage so passing an unregistered language is a compile error.

Behavioral change

Drop the hljs.highlightAuto fallback in Markdown.tsx. With ~35 registered grammars instead of ~190, auto-detection accuracy degrades sharply, so unlabeled / unsupported fences now render as plain monospace via Remarkable's default escaping — same path as the existing lang === 'none' short-circuit. A code comment at the deletion site explains the trade-off, and there's a CHANGELOG entry.

What's NOT in this PR

The 35 grammars still ship in one chunk — they're statically imported, not lazily loaded. The chunk topology is the same as before; the chunk is just smaller (~250 KB vs the prior full set). A TODO at the top of app/utils/hljs.ts documents the path to per-grammar lazy loading and the blocker (Remarkable's highlight callback and the preview loaders all call hljs.highlight synchronously).

Tests

app/utils/hljs.spec.ts adds two assertions:

• Scope tripwire — REGISTERED_LANGUAGES must equal an explicit expected list, so any change to the supported set has to surface in code review.
• Text-loader contract — every key of Text.js's LANGS resolves via hljs.getLanguage(...) from the configured module, catching the case where someone adds an extension to LANGS without registering its language.

Test plan

• npm run lint
• npm run test:only — all 99 files / 893 tests pass (incl. the 2 new hljs assertions)
• npm run build — succeeds; verified kotlin / haxe / unregistered grammars no longer appear in build output, while python / bash / json / yaml / typescript are present in the hljs language chunk
• Manual smoke test:
  • Markdown README with fenced json / yaml / python / bash
  • Markdown README with an unsupported fence (e.g. kotlin) — should render as plain monospace
  • Preview a .json / .yaml / .py file
  • Bucket Download code-snippet panel

Greptile Summary

Replaces the full highlight.js default bundle (~190 grammars) with a curated utils/hljs.ts singleton that registers exactly 35 languages, reducing the hljs chunk size significantly. Seven files are updated only to swap the import path; Markdown.tsx additionally drops the highlightAuto fallback (intentional and well-documented), and Code.tsx's hl prop is narrowed from string to the new RegisteredLanguage union type.

• A "scope tripwire" test locks down the registered-language list and a contract test verifies all Text.js LANGS keys resolve via hljs.getLanguage — but there is no assertion that each entry in REGISTERED_LANGUAGES itself is actually registered.
• The lang === 'none' special case removal in Markdown.tsx is safe: 'none' is not registered, so getLanguage('none') already returned undefined and fell through to return '' in the old else-branch.

Confidence Score: 4/5

Safe to merge; the behavioral change (dropping highlightAuto) is intentional and well-documented, and all existing call sites use languages that are in the registered set.

The implementation is clean and test coverage is good for the primary consumer path (Text.js LANGS). The one gap is that the spec does not assert every entry in REGISTERED_LANGUAGES resolves via hljs.getLanguage, so the exported TypeScript type and the actual runtime registrations can drift silently in future PRs.

catalog/app/utils/hljs.spec.ts — missing assertion that every REGISTERED_LANGUAGES entry is actually registered in the hljs instance.

Important Files Changed

Filename	Overview
catalog/app/utils/hljs.ts	New shared module: imports highlight.js core and registers exactly 35 languages. REGISTERED_LANGUAGES array (source of the RegisteredLanguage type) and registerLanguage calls are maintained in parallel without a guard verifying they stay in sync.
catalog/app/utils/hljs.spec.ts	Adds two tests: a snapshot tripwire for REGISTERED_LANGUAGES and a LANGS-key resolution check. Does not verify that every REGISTERED_LANGUAGES entry resolves via hljs.getLanguage, leaving a gap where the array and actual registrations can drift.
catalog/app/components/Markdown/Markdown.tsx	Swaps highlight.js default import for utils/hljs; drops highlightAuto fallback and the lang==='none' short-circuit, both intentionally. Behavior for unknown/empty lang is unchanged.
catalog/app/containers/Bucket/Download/Code.tsx	hl prop narrowed from string to RegisteredLanguage; all existing callers pass 'python' or 'bash', both in the registered set.
catalog/app/components/Preview/loaders/Text.js	LANGS map exported so hljs.spec.ts can use it as the source of truth for the contract test; no other logic changes.

_{Reviews (1): Last reviewed commit: "Merge remote-tracking branch 'origin/mas..." | Re-trigger Greptile}

Greptile also left 1 inline comment on this PR.

[codecov]

Codecov Report

❌ Patch coverage is 94.87179% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 46.64%. Comparing base (91b30b4) to head (62b5e08).

Files with missing lines	Patch %	Lines
catalog/app/components/Markdown/Markdown.tsx	0.00%	1 Missing ⚠️
catalog/app/containers/Bucket/Download/Code.tsx	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4883      +/-   ##
==========================================
+ Coverage   46.55%   46.64%   +0.09%     
==========================================
  Files         832      833       +1     
  Lines       34131    34162      +31     
  Branches     5833     5831       -2     
==========================================
+ Hits        15890    15936      +46     
+ Misses      16239    16227      -12     
+ Partials     2002     1999       -3     

Flag	Coverage Δ
api-python	93.14% <ø> (ø)
catalog	21.74% <94.87%> (+0.17%)	⬆️
lambda	96.63% <ø> (ø)
py-shared	98.18% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[fiskus]

Wait for #4884 and then maybe implement async parsing.