feat: add SQLite schema adapter

[wondr-wclabs]

Summary

This adds the first database-adapter slice for #4: keep Postgres as the reference implementation, introduce backend selection by DSN, and add SQLite support without adding any new runtime dependency.

I intentionally kept MySQL out of this PR. SQLite is available through the Python standard library, so it exercises the adapter shape without forcing a driver decision. MySQL can now be added as a follow-up optional-extra implementation against the same Schema/Table/Column/ForeignKey model.

What changed

• Add SQLiteDatabase and make_database() so sqlite:///local.db routes through a SQLite backend while existing Postgres DSNs keep the current Database path.
• Open SQLite file-backed databases with mode=ro and also enable PRAGMA query_only = ON, preserving the two-layer read-only model used by Postgres.
• Dispatch schema introspection by database dialect:
  • Postgres keeps the existing pg_catalog mapping.
  • SQLite reads tables/views from sqlite_schema, columns from PRAGMA table_xinfo, and foreign keys from PRAGMA foreign_key_list.
• Pass the selected dialect into sqlglot validation instead of hard-coding Postgres.
• Document SQLite usage and the current scope in README/ARCHITECTURE.

Validation

• .venv/bin/python -m pytest tests/test_schema_adapters.py tests/test_safety.py -q -> 26 passed
• .venv/bin/python -m pytest -q -> 55 passed
• .venv/bin/python -m promptquery --help
• git diff --check