feat: rewrite check-research for Gemini Deep Research (Task 5)

[codercatdev]

Summary

Rewrites app/api/cron/check-research/route.ts to replace all NotebookLM calls with Gemini Deep Research API + Gemini Infographics (Imagen 4 Fast).

What Changed

Removed

• All NotebookLM imports (NotebookLMClient, initAuth, ArtifactTypeCode, ArtifactStatus)
• initAuth() call and NotebookLMClient instantiation
• nbClient parameter from all step functions
• DEFAULT_INFOGRAPHIC_INSTRUCTIONS constant (now lives in gemini-infographics.ts)
• Auth-gated infographic download logic (Imagen returns base64 directly)
• All nbClient.* method calls (pollResearch, importResearchSources, getSummary, getSourceIds, generateInfographic, listArtifacts, getInfographicUrl, getCookieHeader)

Added

• pollResearch(interactionId) from @/lib/services/gemini-research — replaces NotebookLM polling
• parseResearchReport(topic, report) from @/lib/services/gemini-research — replaces manual briefing/summary extraction
• generateInfographicsForTopic(topic, briefing) from @/lib/services/gemini-infographics — replaces NotebookLM infographic generation + download
• researchInteractionId field in PipelineDoc (new Gemini field, with researchNotebookId kept for backward compat)
• enableDeepResearch toggle — when disabled, researching docs skip directly to enriching

Step Changes

Step	Before	After
stepResearching	Polls NotebookLM, imports sources, gets summary	Polls Gemini interaction, parses report with parseResearchReport
stepResearchComplete	Starts async NotebookLM infographic generation → infographics_generating	Generates infographics synchronously via Imagen, uploads to Sanity → enriching
stepInfographicsGenerating	Polls NotebookLM artifacts, downloads auth-gated PNGs	Legacy migration handler only — advances stuck docs to enriching
stepEnriching	Unchanged	Updated buildResearchPayload to handle both new (full ResearchPayload) and legacy (briefing + sources) formats

Kept Unchanged

• fetchCache = 'force-no-store' and maxDuration = 60
• Fail-closed CRON_SECRET auth
• flagStuckDocs() and buildStuckThresholds()
• MAX_DOCS_PER_STATUS = 2
• All helper functions: extractTalkingPoints, extractCodeExamples, classifyScene, classifySourceType
• buildEnrichmentPrompt, claudeCritic, SYSTEM_INSTRUCTION_FALLBACK
• Error handling pattern (try/catch per doc, collect results)

Verification

• ✅ npx tsc --noEmit — zero errors
• ✅ No NotebookLM imports remain (only a comment in legacy migration handler)
• ✅ New Gemini imports present
• ✅ Fail-closed auth verified
• ✅ File reduced from 1006 → 881 lines

Dependencies

• @/lib/services/gemini-research (Task 3 — merged)
• @/lib/services/gemini-infographics (Task 4 — merged)

[vercel]

Deployment failed with the following error:

Resource is limited - try again in 8 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/alex-pattersons-projects?upgradeToPro=build-rate-limit