test(infra) family-wide scraper-pipeline test infrastructure

[jkeeley2073]

Summary

Closes the recurring ⚠️ finding from /local-review runs across PRs #38 / #39 / #40: "no scraper-pipeline integration test asserts yield order, provenance propagation, per-page failure isolation, or the polite-gate routing invariants."

Two shared test fakes plus a proof-of-concept against CgcGamePageScraper. The fakes are reusable across all 8 scrapers; backfilling each one is intentionally deferred to follow-up PRs (so each can land independently as scrapers are touched). What this PR establishes is the template.

What's added

• FakePolitenessGate — implements IPolitenessGate, records every Acquire/Report/lease-dispose so tests can assert the polite-scraping invariants are honored end-to-end. ThrowOnAcquire / ThrowOnReport for testing abort paths.
• QueueingHttpMessageHandler — maps absolute URLs to pre-canned responses. Throws UnexpectedRequestException (with the mapped-URL list embedded in the message) on unmapped requests so a regression that fetches the wrong URL fails loudly instead of silently 404-ing.
• CgcGamePageScraperTests (proof-of-concept) — 5 tests covering:
  • Happy path: yield order, full provenance assertions, URL-equality between gate and wire
  • Per-page failure isolation with politeness invariants still holding under failure
  • Discovery failure aborts the source only (yield-break, no throw)
  • PolitenessException propagation on the Acquire path with assertions that zero requests fired
  • PolitenessException propagation on the Report path

Picked CGC for the proof-of-concept because it exercises BOTH .Game and .Link yields — the template covers the dual-yield case, so backfills to single-yield scrapers (PB, JJP, BoF, Multimorphic, Spooky) will be a strict subset.

Pre-push self-audit (per PR #34 + PR #36)

Step 0 — /local-review

0 🔴 / 5 ⚠️ / 4 ✅ — and all 5 ⚠️ fixed in this PR rather than deferred. The reviewer made a strong asymmetric-cost argument: this PR is the template for ~7 future backfill PRs, so template gaps multiply. Fixing once now is much cheaper than fixing 7 times later.

#	Finding	Action
1	URL-equality assertion between handler.Requests and gate.Acquired/Reported missing — without it a future re-canonicalisation refactor could silently throttle a different origin	Fixed: added Assert.Equal(...) on URL sequences in the happy-path test
2	Per-page failure test threw away gate + handler — politeness invariants under failure not pinned	Fixed: added invariant assertions on the failure path; the 500 response must still be reported back so the 429-streak detector sees real failures
3	PolitenessException test asserted propagation but not "zero requests fired" — a regression that swallowed the throw and continued would still pass for the first call	Fixed: added Assert.Empty(handler.Requests) and Assert.Empty(gate.Reported)
4	Provenance fields (Source.ScrapedFrom, ScrapedAt, DiscoveredOn) not asserted — these are what survive into catalog.json and Phase 2 RAG citations; provenance is "sacred" per CLAUDE.md	Fixed: added explicit assertions on firstGame.Source!.ScrapedFrom etc.
5	ThrowOnReport was unused API surface ("I added a hook, never tested it")	Fixed: added a 5th test pinning the Report-side throw path

Step 1 — Mechanical checklist

• No new *Options properties (N/A)
• No scraper changes (sibling-diff N/A)
• No bare catch { } in tests/PinballWizard.Scraper.Tests/Scraping/_TestInfra/
• SourceAliasContractTests still passes
• Tests assert behavior — every test pins a specific invariant the unit-test surface couldn't reach
• Build is zero-warning
• git log -1 --format='%an <%ae>' — personal noreply

Tests

378 / 378 passing (was 373). +5:

• CgcGamePageScraperTests (5)

Folder convention

Established new convention: tests/.../Scraping/_TestInfra/ for shared test fakes. Leading underscore sorts to top of directory listings — discoverability for the next test author. (Reviewer noted this is a new convention; one-line README in a follow-up could document it formally if it sticks.)

Out of scope

• Backfill across the other 7 scrapers — separate PRs each, can land independently. Each backfill should copy the 5-test pattern from CgcGamePageScraperTests. Single-yield scrapers (PB / JJP / BoF / Multimorphic / Spooky) drop the Link-yield assertions; multi-yield scrapers (AP / Stern's GamePageScraper) keep them.
• README in _TestInfra/ — defer until pattern proves out across 2-3 backfills.

🤖 Generated with Claude Code