fix: stop dropping token labels with more than 100 rows

[dawsbot]

Problem

scripts/HtmlParser/HtmlParser.ts had a long-standing TODO ("repull tokens which have a page length > 100"). selectAllLabels skipped any token label whose record count was >= 100 (the token page-size cap), so those categories were silently never pulled at all — in the etherscan labelcloud fixture alone that is 35 labels, including defi (2,101 tokens) and stablecoin. The URLs it did emit also lacked the start cursor that ApiParser.fetchTokens requires, which would crash that fetch path.

Fix

• HtmlParser.selectAllLabels now always includes token labels, emitting ?size=100&start=0 — matching CheerioParser.selectAllLabels, the variant ChainPuller uses at runtime.
• Pagination happens in the existing fetch layer: ApiParser.fetchTokens keeps advancing start by 100 until a page comes back with fewer than 100 rows, so all rows are captured.
• The accounts guard (skip labels with >= 10,000 records to avoid freezing on e.g. beacon-depositor) is unchanged.

Tests

• New ApiParser unit test drives fetchTokens through a full 100-row page plus a 20-row page via a stubbed browser fetcher (rows built from the real aave.json fixture HTML with unique addresses), asserting the start cursor advances [0, 100] and all 120 unique rows are captured.
• Labelcloud fixture tests extended: previously-dropped >100-row token labels (defi on etherscan/arbiscan/optimism, blockchain on bscscan) are now asserted present with the pagination cursor, and expected label counts updated (etherscan 899 → 934, bscscan 240 → 263, arbiscan 167 → 170, optimism 101 → 102).

bun test:unit (31 pass), eslint, and tsc all pass. Dataset intentionally not regenerated — code fix + tests only.

🤖 Generated with Claude Code