[Docs] Update the ARCHITECTURE.md, METHODOLOGY.md, etc. files to reflect current architecture

[nojibe]

Description

Realigns ARCHITECTURE.md, METHODOLOGY.md, BLUEPRINT_FORMAT.md, and INTER_AGREEMENT_PLAN.md with the current system. Several diagrams and sections had drifted from the code.

Summary

ARCHITECTURE.md

• Automated evaluations now attributed to GitHub Actions + Railway (was Netlify), in both the diagram and the "Automated Workflow Components" section.
• Storage diagram: adds workshop/, pr-evals/, and ndeltas//vibes//compass/ under live/models/, with descriptions.
• Documents the previously undocumented api/internal/ background routes (execute-pr-evaluation-background, execute-api-evaluation-background, generate-pairs-background, …).
• Interactive workflow diagram: splits sandbox polling into its own Status API node, corrects S3 path /sandbox-runs/ → live/sandbox/runs/, corrects the dispatch edge label — callBackgroundFunction awaits a response up to a 30s timeout; Railway
  continues running the handler to completion if that window is exceeded.
• Pipeline diagram: replaces "5-point scale" with "10-class experimental ordinal scale"; links METHODOLOGY.md.
• Fixes search_index.json → search-index.json (matches SEARCH_INDEX_FILENAME constant in storageService.ts).
• Clarifies that calculateHybridScore is defined in calculationUtils.ts; summaryCalculationUtils.ts orchestrates it, not defines it.
• Removes redundant prose: one-line descriptions that repeated adjacent headings, and a core.json lazy-loading paragraph that duplicated the Section 4 architectural concept.
• Fixes markdownlint violations (MD022, MD007, MD032): blank lines around headings/lists, 2-space list indent.

METHODOLOGY.md

• Updates the experimental classification scale to the current 10-class ordinal.
• Krippendorff's α note now references the active scale instead of hardcoding the old values.

BLUEPRINT_FORMAT.md & INTER_AGREEMENT_PLAN.md

• Corrects useExperimentalScale description from "9-point" to "10-class"; notes that FORCE_EXPERIMENTAL = true currently makes the flag a no-op.
• Refreshes default-judges example (stale qwen/GPT-OSS models → current set); adds cross-family consensus rationale.
• Fixes backup-judge trigger condition ("any" primary judge fails, not "all") and corrects the model identifier.
• Fixes stale Netlify Blobs reference in INTER_AGREEMENT_PLAN.md — storage uses storageService (S3 in production, local FS in dev).

Testing

• Cross-referenced every changed claim against source: workflow files in .github/workflows/, handlers in app/api/internal/, and S3
  path constants in storageService.ts.
• Re-ran markdownlint locally; no remaining violations.
• Previewed both docs on GitHub to confirm diagrams and anchor links render.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.