Fix CI python version mismatch, address minor review findings, add pg…

[aersam]

…devkit.db and fetch-missing

• CI: bump the workflow's python-version matrix to 3.14 to match requires-python, bump checkout/setup-python actions, and install podman (needed by conftest.py's ensure_testdb() on ubuntu-latest).
• schema.py: generalize cross-file dependency tracking beyond tables so views/functions/procedures referencing each other also get correct apply ordering, not just table FKs. Also fix apply_schema's retry pass to re-seed .test_data.json, not just re-run the raw SQL.
• query.py: replace naive ';'-splitting with a dollar-quote/string-literal aware statement splitter so run-sql on a file with a plpgsql function body doesn't get cut in half.
• container.py: add test coverage for the "name already in use" race fallback branch.
• config.py: validate extensions is actually a list.
• tests/testdb: give test project names a per-process suffix so concurrent pytest runs (e.g. from separate worktrees) against the shared container don't drop each other's databases.
• Add pgdevkit.db: an importable psycopg CRUD helper module (PgPool, PostgresTableModel, SqlLoader, pg_retrieve/pg_insert/pg_upsert/...) so projects depend on a version instead of copy-pasting reference files.
• Add pgdb fetch-missing: reverse-engineers DDL for tables/views/functions that exist in the database but aren't tracked under database/, reusing the existing introspect/diff machinery.
• Add docs/database-layout.md as the single source for the database/ folder convention, and a consolidated skills/pgdevkit/SKILL.md superseding postgres-test-setup, postgres-best-practices, and database-in-source (left in place in the separate skills repo for now).

Claude-Session: https://claude.ai/code/session_01GpwZzpmtp1c1ZHp9LT1Asd