docs: architecture/sequence/module diagrams, real CLI screenshots, badges, CI

[codeapplied]

Summary

• docs/ARCHITECTURE.md — data-flow diagram (including the needs_review dead-end path), a full pipeline sequence diagram, a components table, a module-dependency diagram, and the reasoning behind every key design decision (confidence gate, field-level authority, true dry-run safety, department-priority ranking, per-project leads, the overwrite gate, sandbox-first vendors, naive UTC)
• docs/ADDING_A_VENDOR.md — how to wire in a real search vendor, contact-enrichment vendor, or CRM, mirroring each interface's existing template/real-client pattern
• .github/workflows/tests.yml — CI running the 46-test suite on push/PR to main (mirrors the sibling tender-tracking-system repo's own workflow). No scheduled daily-run workflow yet — deliberately skipped since there's no real vendor wired in yet, so a scheduled run would only touch sandbox data
• scripts/generate_readme_assets.py — regenerates real CLI-output SVGs (not mockups) by running the actual CLI against a throwaway temp DB
• README: badges (Tests/Python/License), a compact top-level architecture diagram + link to the full doc, an "In action" section with 4 real CLI screenshots, a dedicated Testing section, trimmed the three module deep-dive sections down to summaries (full detail now lives in ARCHITECTURE.md instead of being duplicated in both places)

Test plan

• pytest — 46 tests still passing after the README/docs changes (no code changes in this PR)
• Ran scripts/generate_readme_assets.py for real — verified all 4 SVGs were written with real content (not empty/broken), and that the underlying CLI output matches what's documented (2 of 5 sample orgs resolve at high confidence, the aggregator false-positive is correctly excluded leaving a low-confidence single-source match, push-crm dry-run correctly shows the pre-seeded org reused via dedup)

🤖 Generated with Claude Code