Tracking Issue: own infrastructure for duckdb builds

[myrrc]

Currently we use duckdb servers to download libduckdb.so and headers.
This has some drawbacks:

1. Servers sometimes are unavailable, and our builds fail.
2. We don't have libduckdb.so with debug symbols, so we always need to rebuild locally which is a big waste of time.
3. For pre-releases, we can't use duckdb servers because there are no pre-release builds, and we want to test changes before release.

So we need an S3 bucket with release/pre-release .so files for architectures we use as well as headers, and CI infrastructure to update it automatically.