Sample masking from sgkit dataset

[benjeffery]

To restrict inference to a subset of samples a subset-copy of the sgkit dataset needs to be performed. For large datasets this is less than ideal.
If tsinfer read a sample_mask array from the dataset, inference could be restricted to those samples in the mask. If done at the SgkitSampleDataset level this should be possible without wider changes to tsinfer.

[benjeffery]

Note that "sample" in this context is the sgkit sample, which is "individual" in tsinfer.