refactor: restructure project for production readiness

[spideystreet]

Summary

Full restructure of the OST Linker codebase (278 commits, 190 files changed) covering every layer of the stack:

• Pipeline & Dagster — complete asset graph with ingestion, classification, embedding, matching, and sync groups; jobs, schedules (5×/day), and sensors; custom PandasPostgresIOManager
• dbt — staging → intermediate → marts model hierarchy with proper naming (stg_, int_, fct_, match_); SQLFluff-clean SQL; schema tests and docs
• Go services — scraper and fetcher binaries invoked as subprocesses by Dagster assets
• ML — SentenceTransformer embeddings (384-dim, MiniLM-L6-v2) + FastText language detection + LLM classification via OpenRouter
• Database — Prisma schema across 4 PostgreSQL schemas (public, github, ml, match) with pgvector
• Docker — 3-stage build (Go builder → Python builder → runtime), CPU-only torch (~2GB savings), non-root user, healthcheck
• CI/CD — reusable quality-checks.yml with lint, format, type check, unit tests (80% coverage), Go vet+build, Docker build, Prisma validate, pip-audit, gitleaks, dbt parse, docs submodule validation
• Docs — git submodule pointing to ost-docs, automated sync workflow to create PRs on ost-docs when docs change
• Repo hygiene — cleaned .gitignore, untracked 131MB FastText binary, tracked utility scripts, concise README

Test plan

• ruff check src/ — no errors
• ruff format --check src/ — formatted
• mypy src/ — passes strict mode
• pytest -m unit — 20/20 tests, 100% coverage
• go vet + go build — both scraper and fetcher compile
• prisma validate — schema valid
• pip-audit — no vulnerabilities (dbt-common upgraded to 1.37.3)
• docker build — image builds successfully with CPU-only torch
• CI pipeline runs on PR creation

🤖 Generated with Claude Code