Add async/asyncio support

## Summary
`crate-python` is entirely synchronous. Every call to `Connection`, `Cursor.execute()`, `fetchall()`, and the `BLOB API` blocks the calling thread. This makes the driver unusable in modern async Python stacks like FastAPI route handlers, Django async views, Starlette/Litestar, asyncio-based ETL pipelines, and LangChain/LlamaIndex tool integrations that all expect await-able database calls.

This issue tracks the design and implementation of a first-class async interface for `crate-python`.

## Motivation
New support would increase use in:
- AI/ML backends: FastAPI services that call both a vector store and CrateDB in the same request handler
- Real-time dashboards: async WebSocket handlers that need concurrent DB queries
- High-throughput ingest: asyncio pipelines where thread-per-connection doesn't scale

Start by mapping the existing Connection, Cursor.execute(), fetchall(), and BLOB API entry points to understand the synchronous surface. Review how these calls are used in the current driver and define what the first-class async interface must cover. Done means async Python callers can use awaitable equivalents without blocking the calling thread.

Written by the indexing model from the issue text.