Allow benchmarking larger-than-memory datasets

[ashkrisk]

Build on patches introduced earlier in the series to allow BenchYAML to process that don't fit in physical memory, by memory-mapping the index vectors instead of fully loading them upfront.

This behaviour is opt-in and requires you to set -Djvector.bench.dataset.mmap.enable=true to enable.

Additionally, fixes a concurrency issue with RAVV handling in Grid, and makes the output directories configurable

Summary of changes:

• Add a RavvDataSet wherein the index vectors are backed by a RAVV.
• Add a function to DataSetInfo to return a DataSet backed by memory-mapped files, using RavvDataSet.
• Consume the new DataSet from BenchYAML
• Fix a concurrency issue in Grid where RAVV#getVector was being called from multiple threads at once
• Make Grid's work directory and index cache directory configurable via environment variables. (Useful if you have an SSD on a separate mount)

---

[github-actions]

Before you submit for review:

• Does your PR follow guidelines from CONTRIBUTIONS.md?
• Did you summarize what this PR does clearly and concisely?
• Did you include performance data for changes which may be performance impacting?
• Did you include useful docs for any user-facing changes or features?
• Did you include useful javadocs for developer oriented changes, explaining new concepts or key changes?
• Did you rebase your branch onto the latest main for regression testing and PR submission?
• Did you trigger regression testing via Run Bench Main and review results?
• Did you adhere to the code formatting guidelines (TBD)
• Did you group your changes for easy review, providing meaningful descriptions for each commit?
• Did you ensure that all files contain the correct copyright header?
• Did you add documentation for this feature to the release notes directory?

If you did not complete any of these, then please explain below.

[jshook]

Why not just make getDataSet() return a DataSet, and have the implementation of that which is provided support RAVV? That was always the intent here.

A DataSet may implement its own loading methods by design. Some of our current implementation have historic limitations which should be removed. The way to do this would be to have a DataSet impl type which does it properly, and to remove any other surface but the RAVV style accessor.

To keep that supported, the getDataSet on DataSetInfo should remain a contract type, not any concrete type.

[jshook]

Let's align on the contract vs concrete boundaries.