Add Cloudflare Hyperdrive support for Python Workers

[CollierKing]

Summary

• Adds create_engine_from_hyperdrive() and hyperdrive_connection() for using SQLAlchemy against PostgreSQL through a Cloudflare Hyperdrive binding inside a Python Worker
• No custom dialect is involved: Hyperdrive speaks the PostgreSQL wire protocol, so SQLAlchemy's own postgresql+pg8000 does the work and this package supplies only the binding-to-engine glue
• Engines use NullPool, since a Worker cannot reuse sockets across requests and Hyperdrive pools server-side
• hyperdrive_connection() holds a per-event-loop asyncio.Lock, because the Workers socket layer does not tolerate concurrent synchronous driver I/O
• Driver is auto-detected from the binding's scheme, falling back to connectionString, and can be overridden with driver=
• Adds a [hyperdrive] extra and an examples/workers-hyperdrive/ example Worker
• Adds 22 unit tests and 6 Worker integration tests; bumps to 0.4.0

Driver support

Cloudflare documents five drivers for Python Workers. Only two can back a SQLAlchemy engine, verified by running each from inside a Worker:

Driver	Status	Notes
pg8000	Supported	PostgreSQL, covered by integration tests
pymysql	Best effort	MySQL, verified manually against MySQL 8, no automated coverage
psycopg	Unsupported	Needs libpq, absent from the Workers Pyodide build
asyncpg	Unsupported	Async-only
aiomysql	Unsupported	Async-only

Only synchronous SQLAlchemy works. asyncpg and aiomysql run fine in Workers on their own, but driving them from SQLAlchemy requires create_async_engine() and therefore greenlet, which Workers lack — so the two drivers Cloudflare recommends are precisely the two that cannot be used here.

Both supported drivers need TLS explicitly disabled, spelled differently for each, since Hyperdrive terminates TLS itself.

Notes for reviewers

• Hyperdrive has no local emulation. wrangler dev always connects to the Postgres named by CLOUDFLARE_HYPERDRIVE_LOCAL_CONNECTION_STRING_HYPERDRIVE, and "remote": true is not valid on a hyperdrive binding. Integration tests skip unless that variable is set.
• Behaviour was additionally confirmed against a real deployed Worker, which is the only way to exercise the actual Hyperdrive config rather than the local proxy.
• The integration fixture creates its table before the run and drops it afterwards, since a Hyperdrive config often points at a shared database.
• The example Worker's schema-changing endpoints are gated behind an ADMIN_TOKEN when one is set, because a deployed Worker is publicly reachable.
• __version__ had drifted to 0.3.1 while the package shipped 0.3.11. The version is now derived from __version__ via hatchling, with a test asserting the two agree.

Testing

• 43 unit tests pass on Python 3.9, 3.10, 3.11 and 3.12
• 6 new Hyperdrive Worker integration tests pass against PostgreSQL 17
• 66 existing D1 Worker integration tests still pass
• ruff check, ruff format --diff, codespell, mypy src/, pre-commit --all-files and uv build all clean