feat: inline cross-reference hyperlinks in law reader

[daffaromero]

Summary

• Adds Wikipedia-style inline cross-reference links to pasal body text in the law reader
• Intra-document "Pasal N" / "Pasal N ayat (M)" references become anchor links that scroll to the target pasal and trigger the existing `HashHighlighter` green flash
• Cross-document UU/PP/Perpres/Perppu/Perda citations are resolved server-side against a `worksLookup` map and rendered as `` components navigating to the correct law page; unresolvable citations fall back to plain text (no broken links)

Note: This branch is rebased onto `fix/structured-law-pasal-loading` (PR #21) to avoid a merge conflict and regression. Either order of merging is now safe — both PRs agree on the `!hasStructure` pagination guard.

Changes

File	Change
`src/lib/crossref.ts`	Pure tokenizer — parses legal citation patterns into `Token` discriminated union
`src/lib/tests/crossref.test.ts`	20 Vitest tests covering all token types, edge cases, Perpu/Perppu aliases, Perpres, Perda, optional Nomor keyword, mixed references, and unknown types
`src/components/reader/RichPasalContent.tsx`	`"use client"` component that renders tokenized text as links or plain fragments
`src/app/[locale]/peraturan/[type]/[slug]/page.tsx`	Adds `worksLookup` map (built server-side from all works, passed to reader); preserves `!hasStructure` pagination guard from PR #21
`src/components/reader/PasalList.tsx`	Threads `worksLookup` prop down to `PasalBlock`
`src/components/reader/PasalBlock.tsx`	Replaces plain text ` ` with `RichPasalContent`

Bugs fixed

Bug	Fix
`citationToSlugKey()` emitted `uu-no-13-tahun-2003` but DB trigger `generate_work_slug()` produces `uu-13-2003` — cross-document links were silently falling back to plain text 100% of the time	Fixed `citationToSlugKey()` to emit `{type}-{number}-{year}` format; updated test fixtures to match
Scanned PDFs use all-caps `UNDANG-UNDANG` / `PASAL` — not matched by original regex	Added `i` flag to `CROSSREF_RE` (`gi`), making the entire outer regex case-insensitive

Test Plan

• Open a law page with cross-references (e.g. `/peraturan/uu/uu-1-2026`)
• Verify "Pasal N" text renders as a dotted-underlined link
• Click a pasal link — should scroll to anchor and flash green via `HashHighlighter`
• Verify a cross-UU citation (e.g. `UNDANG-UNDANG Nomor 6 Tahun 2023`) renders as a link to `/peraturan/uu/uu-6-2023`
• Verify a citation for a law not in DB renders as plain text (no broken link)
• Verify a structured law with 100+ pasals (e.g. UU Cipta Kerja) still renders all BABs correctly (pagination guard intact)
• 44/44 Vitest tests passing (20 crossref + 24 across other test files)
• No new lint errors introduced

[daffaromero]

PR Description Audit

Went through the description against the actual code. A few things to correct:

---

1. Test count is wrong — 20, not 31

The description says "31/31 Vitest tests passing". The test file (src/lib/__tests__/crossref.test.ts) has 20 it() blocks, not 31. Easy to miscount if you're going by describe blocks rather than individual cases.

---

2. Cross-document UU links are actually still broken (slug format mismatch)

This is the most significant issue. The "Bugs fixed" table says:

Cross-UU slug was uu-13-2003 but DB slugs are uu-no-13-tahun-2003 | Fixed citationToSlugKey() to emit uu-no-${number}-tahun-${year}

This is inverted. The DB trigger (generate_work_slug() in migration 053) produces slugs in the format:

{type_prefix}-{number}-{year}

e.g. "uu-13-2003", "pp-74-2008". There is no no- or tahun- in DB slugs.

The worksLookup map is built from w.slug directly (line 251–258 of page.tsx), with the comment even confirming: "uu-13-2003" → "/peraturan/uu/uu-13-2003".

But citationToSlugKey("Undang-Undang Nomor 13 Tahun 2003") returns "uu-no-13-tahun-2003". That key will never exist in worksLookup, so worksLookup[slugKey] is always undefined, and all cross-document citations silently fall back to plain text.

The tests pass because they construct the worksLookup fixture with the uu-no-13-tahun-2003 key directly — they test that the tokenizer resolves a key it's given, not that the key matches what the DB actually produces.

The actual correct fix would be to change citationToSlugKey to emit "uu-13-2003" format (matching the DB trigger), or to build the worksLookup map using a derived key in uu-no-N-tahun-YYYY format alongside the raw slug. Either way, the formats need to agree.

---

3. Bug fix description for the i flag is inaccurate

The table says:

Scanned PDFs use capital-A Ayat — not matched by original regex | Added i flag to all inner regexes

Looking at the code, the i flag was added to CROSSREF_RE itself (the outer regex, making it "gi"). The pasal match group (?:Pasal) is now case-insensitive via the outer flag. There are no "inner regexes" that had i flags added — the numMatch extractor inside the tokenizer already matched case-insensitively via the outer i. The description over-complicates what actually happened.

---

4. Minor: worksLookup ISR framing is slightly off

The description says "Fetches worksLookup map server-side (ISR 24h)". The ISR 24h comes from export const revalidate = 86400 at the page level — it applies to the whole page, not specifically to the worksLookup query. The query itself is just one of the parallel fetches inside LawReaderSection, with no dedicated cache. Not wrong per se, but the phrasing implies the lookup has its own separate caching strategy.

---

Summary

Claim	Status
31 Vitest tests	Incorrect — 20 tests
Cross-UU slug bug "fixed"	Incorrect — still broken (format mismatch between citationToSlugKey and DB slugs)
i flag added to "all inner regexes"	Inaccurate — added to outer CROSSREF_RE only
worksLookup ISR 24h	Slightly misleading framing