Add full X-chromosome example dataset as VCF Zarr

[andrewkern]

Closes #161.

examples/scikit_allel_comparison.py defaults to examples/data/gamb.X.phased.n100.zarr, which was never tracked in the repo. .gitignore excludes data/, and only the 4 Mb subset store was ever force-added. A fresh clone could run the example with --small and no other way.

Rather than commit the original store as-is, this ships it re-encoded with bio2zarr. The original was a scikit-allel-era Zarr v2 store carrying all 26 VCF fields at full 5.3M-variant length with 1000-variant chunks, which is what produced 90,126 files for 55 MB of data.

Two changes: drop the ten fields pg_gpu/zarr_io.py never reads (call_PQ, call_PS, variant_AA, variant_AAProb, variant_AAcond, variant_AC, variant_AF, variant_AN, variant_CM, variant_quality), and raise the variant chunk size from 1000 to 100000.

	original store	this PR
Data	55 MB	15 MB
Checkout footprint	356 MB	16 MB
Files	90,126	411
Git objects	47,878	220
Added to pack	52.8 MiB	14.0 MiB
Load time	20.3 s	7.9 s

For scale, the 4 Mb subset store already in the repo is 17,400 files; this full chromosome is 411.

Verification

Loaded both stores and compared:

shape     identical=True   (200, 5297731)
samples   identical=True   ['AA0045-C', 'AA0046-C']
pos0/posN identical=True   3 / 24392613
pi        identical=True   0.0021073932663811298
theta_w   identical=True   0.008740797101718526
tajimasD  identical=True   -2.4556936206779456

The example itself runs end-to-end against the new store, with all of its internal cross-checks against scikit-allel passing:

Loading gamb.X.phased.n100.vcz ...
  loaded in 8.50s; 200 haplotypes, 5,297,731 variants, chrom range 3-24,392,613
  pi:        max abs diff = 3.339e-17
  theta_w:   max abs diff = 1.492e-16
  tajimas_d: max abs diff = 3.286e-14
  median r^2: max abs diff = 1.514e-08

Notes

• The file is renamed .zarr to .vcz to match the bio2zarr convention; the example's path constant is updated to match.
• The exact vcf2zarr commands that regenerate the store are recorded in a comment at the top of the example, so it is reproducible from the VCF already in examples/data/.
• The missing-file error text told users to run git lfs pull, but this repo does not use LFS. Corrected to point at the regeneration commands.
• The same treatment would shrink the 4 Mb subset store considerably as well; left out of scope here.