Add CAGRA build heuristics independent of HNSW parameters

[achirkin]

Split the cagra::index_params::from_hnsw_params into a core cagra::index_params::from_dataset and a thin wrapper ``cagra::index_params::from_hnsw_params`. This solves two problems:

1. Avoid using the HNSW parameters that do not make sense for CAGRA
2. Allow HNSW parameter heuristics change without affecting CAGRA in future

The new parameter defaults are selected to yield exactly the same configurations at old call sites of from_hnsw_params

[tfeher]

Thanks Artem, it makes sense to separate the dataset shape dependent CAGRA heuristics to general CAGRA heuristic and HNSW specific ones. The PR looks good to me, apart from two small docstring issues.

[cjnolet]

This is a relatively small PR in comparison to the dataset changes (that we absolutely need merged for this release).

If we keep merging (lower priority) changes that depend on the current dataset APIs, we aren't going to be able to merge the datasets api changes in time for the release.

Let's at least wait until @HowardHuang1's PR is merged for this one. Given how close it is to the release, I'd almost suggest we proactively push this to 26.10.

[achirkin]

@cjnolet I understand, this PR surely can wait.
FIY, it shouldn't overlap with the dataset API PR in any way, it only adds two functions that consume extents (not an mdspan, mdarray, or a new dataset type): caga::index_params::from_dataset and cagra::index_params::graph_build_heuristic.

[cjnolet]

Thanks @achirkin. I see now, this is expecting the shape information individually, instead of through an mdspan/dataset.

@imotov if not too much trouble, can you merge this in your heuristic PR and work from that? I agree with Artem here that the change seems not too terrible on the Java side. Will help us streamline a bit more.