Vectordata streaming support for examples/bench

[jshook]

This extends the examples/bench capabilities in jvector:

• The DataSet type is virtualized behind an interface
• The way DataSets are loaded is more modular
  • A DataSet loader has been added to support the vectordata API

The net effect is that bench can no access vectordata hosted datasets.
The benefits of this are several:

• remote vector test data hosting
• uniform API for finding and using vector datasets
• merkle-based automatic download of chunks for dynamic access
• efficient download and automatic caching of data locally on test nodes
• mapping and ranging subsets of data, such as "first 1M", "first 10M" and so on under profile names
• packing various vector data views in a consistent API: base, query, ...
• management of datasets via catalogs, orthogonal to access control

Most of the core wiring of these capabilities is provided by another library which is part of the nosqlbench project. The changes to jvector are to adapt it to use these APIs.

There are a couple issues yet to resolve with the Java version configs and GHA.

[jshook]

Due to conflicts with the executable jar approach, I'll need to refactor this a bit, so I moved it to a draft.
My plan for this is to combine the commands so that autobench is included and update the GHA wiring.