Fix Google generated-module detection and continue Notandia runtime rebrand

[jnton]

Why

Manual testing exposed connected browser regressions and an architectural mismatch:

1. Google organic/Scholar results that link through PubMed Central or Europe PMC identifiers were not consistently reaching the same work-identity path;
2. Google AI Overview and People Also Ask are composite answer containers, but earlier selectors could miss their actual source cards or style an entire FAQ/question because one nested source matched;
3. publisher context and formal-integrity context had begun to diverge in how they recognized source/work evidence;
4. the popup quick-settings mode and overview counts had presentation/state inconsistencies;
5. current general-purpose runtime naming still needed to complete the Notandia rebrand without breaking upgrade compatibility.

Source-context contract

Notandia operates on source/citation units, not only scholarly papers. A source may be a Google result, AI Overview source card, People Also Ask source, Wikipedia reference, Healthline trusted source, or journal bibliography entry.

Publisher/watchlist context can use exact URL/domain evidence without requiring a DOI. DOI/PMID/PMCID/arXiv identity is optional enrichment on the same source record and enables DOI-based formal post-publication checks when resolvable.

Changes

• use one NotandiaWorkIdentifiers mapper for DOI/PMID/PMCID/arXiv recognition;
• normalize Europe PMC /article/med/<PMID>, /article/pmc/<numeric-id>, and /article/pmc/PMC<id> URLs in the shared mapper;
• resolve PMID/PMCID evidence through the centralized, NCBI-compliant background provider before DOI-dependent matching/checks;
• remove the obsolete page-fetch interception path so NCBI traffic has one governed route;
• keep ordinary Google results separate from composite modules;
• treat Google AI Overview span.WBgIic cited-source wrappers and li.h7wxwc > div.cRH23c[data-src-id] source cards as source units;
• never treat the outer AI Overview as a source;
• never treat .related-question-pair itself as a source; scan its concrete source pills/cards instead;
• keep dynamic rescanning for sources inserted after expansion;
• make NotandiaReferenceSelectors canonical while retaining the released alias for upgrade compatibility;
• preserve Wikipedia and Healthline citation/source adapters as part of the general source model;
• make NotandiaDomains, NotandiaDomainUtils, NotandiaNcbiApiHandler, NotandiaSettings, data-notandia-ref-id, and data-notandia-doi canonical current runtime names;
• retain legacy runtime names/DOM attributes only as explicit aliases required for upgrade compatibility;
• replace the popup gear glyph with an SVG control and make quick settings a real toggle mode rather than a stacked panel;
• reconcile popup overview/reference counts;
• add regression and policy tests for source-first context, Google source isolation, Europe PMC identities, public Notandia identity, and compatibility boundaries.

Rebrand boundary

MDPI remains valid when it means the publisher, its domains/DOI prefix, or the mdpi publisher profile. Legacy storage/runtime identifiers are removed only through tested migrations so existing Chrome/Edge installations and settings remain intact.

Manual gate

Do not merge or publish this PR until live testing confirms:

• the outer AI Overview is never styled as a publisher;
• matching AI Overview source pills/cards are styled independently;
• expanding AI Overview discovers newly inserted sources;
• the outer People Also Ask module and .related-question-pair wrappers are never styled because of nested matches;
• matching source pills/cards inside expanded questions are evaluated independently;
• ordinary Google organic results still work;
• Google Scholar Europe PMC /article/pmc/7102549 resolves to PMCID PMC7102549 and, with NCBI metadata enabled, to the expected DOI;
• Wikipedia and Healthline source adapters still work;
• popup counts do not inflate because of duplicate/hidden Google source UI.

[jnton]

The underlying publisher-vs-integrity source-detection split has been isolated into draft PR #52. #52 makes one shared source-record pipeline authoritative for both consumers and restores configured search-result records (including Google Scholar div.gs_r) to formal-integrity scanning. Its automated security/regression tests, cross-browser builds, and generated-runtime verification are green.

Recommendation: do not merge #51 as the architecture baseline. Live-test #52 first, then rebase/port #51's still-useful AI Overview / People Also Ask and popup work onto the shared source context rather than maintaining parallel scanner logic.