test(infra) Stern Manuals scraper-pipeline integration tests

[jkeeley2073]

Summary

Backfill of the PR #41 family-wide test-infra template against Stern's ManualsScraper. First non-Game-yielding scraper in the family backfill — manuals are documents, not games, so the tests assert .Link yield order with full provenance only, no .Game items, and no .Link.GameSlug parent-lineage (the scraper does not derive game slugs from filenames).

5 tests using the shared fakes (FakePolitenessGate + QueueingHttpMessageHandler from PR #41):

1. ScrapeAsync_HappyPath_YieldsLinksInPageOrderWithProvenance — three same-host PDFs in page order, plus a non-PDF page link and an off-host PDF that exercise both filter branches; asserts full provenance (SourceType.ManualsPage, DiscoveryUrl, both inner and outer DiscoveryContext) and the gate-vs-wire URL equality invariant.
2. ScrapeAsync_OneAnchorWithMissingHref_DoesNotAbortRun — adapted from the CGC template's per-page-500 test. ManualsScraper has no per-link HTTP fetch (the entire scrape is one page → many .Links parsed inline), so the per-item failure mode is at the anchor-parsing layer. Fixture mixes empty-href, whitespace-href, off-host, and non-PDF rejection cases between two valid PDFs; asserts the loop continues past every filter branch.
3. ScrapeAsync_DiscoveryFailure_AbortsThisSourceOnly — /manuals/ 500s, scraper yields nothing without throwing; politeness invariants still hold (the 500 is reported back, lease disposed).
4. ScrapeAsync_PolitenessExceptionFromGate_PropagatesUp — gate.ThrowOnAcquire throws PolitenessException; asserts propagation AND zero requests / zero reports (the throw fires before any HTTP call).
5. ScrapeAsync_GateThrowsOnReport_BubblesUp — symmetric: gate.ThrowOnReport propagation.

Tests: 430 → 435 (+5). Build clean, zero warnings.

Adaptations from the CGC template

ManualsScraper differs from CgcGamePageScraper in three ways that shape the test surface:

• Single yield kind — .Link only. The happy-path test removes all .Game assertions.
• Single HTTP fetch — only /manuals/ is fetched; there are no per-link HTTP requests. Politeness assertions check Single(handler.Requests) instead of handler.Requests.Count == gate.Acquired.Count.
• GameSlug is not derived from filenames — ManualsScraper leaves DiscoveredLink.GameSlug null. The happy-path test pins this explicitly with a comment noting that cross-linking happens later (or, per the project's known-gap list, doesn't happen at all yet).

The class-level <remarks> block documents these adaptations so a future reviewer can verify the divergence is intentional.

No bugs found in ManualsScraper.

Pre-push self-audit

Step 0 — Local review: Deferred to the human reviewer at merge time per the task brief.

Step 1 — 7-item mechanical checklist:

1. Every option field is read — N/A (no new option fields). PASS.
2. Sibling-diff for drift vs CGC template — adaptations enumerated above and documented in <remarks>. PASS.
3. No bare catch {} — no catch blocks in test file. PASS.
4. CLI / orchestrator wiring is end-to-end — no new ISourceScraper added; existing SourceAliasContractTests already pin ManualsScraper/manuals. PASS.
5. Tests assert behavior, not just structure — happy-path includes filter-rejection anchors so the filter assertions are load-bearing; failure-isolation test mixes four rejection reasons between two valid PDFs; politeness-throw tests assert zero requests/reports to prove the throw fires before HTTP. PASS.
6. Build is zero-warning. PASS.
7. Identity check: Jim Keeley <94459922+jkeeley2073@users.noreply.github.com>. PASS.

Test plan

• dotnet build — 0 warnings, 0 errors
• dotnet test --filter ManualsScraperTests — 5/5 pass
• dotnet test — 435/435 pass (was 430)