Bgen

[SantiDu]

Three changes were made here:

1. Conda/Mamba and Pixi seem to install numpy>=2 even though the yaml file states numpy=1.26, but scipy requires numpy<2:

/home/j/jidu/MetaXcan-0.8.1/software/metax/gwas/GWAS.py:5: UserWarning: A NumPy version >=1.22.4 and <1.29.0 is required for this version of SciPy (detected version 2.0.2)
  import scipy.stats as stats
Traceback (most recent call last):
  File "/home/j/jidu/MetaXcan-0.8.1/software/Predict.py", line 16, in <module>
    from metax.genotype import Genotype
...
    from ._kdtree import *
  File "/scratch/tmp/jidu/predict/.pixi/envs/imlabtools/lib/python3.9/site-packages/scipy/spatial/_kdtree.py", line 4, in <module>
    from ._ckdtree import cKDTree, cKDTreeNode
  File "_ckdtree.pyx", line 1, in init scipy.spatial._ckdtree
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

The proposed change in the yaml file uses pip to install scipy and numpy so this incompatibility disappeared.

2. It takes forever to use bgen-reader to iterate a bgen file, so I switched to the bgen package. Now the speed of using bgen for prediction is really faster than using vcf.

3. Update version from 0.7.5 to 0.8.2.

Please ignore all the commits that try out combinations of dependency package version numbers hahahaha..

[hakyim]

Thanks for the PR! Before merging, one question: did you test with a phased BGEN file? The dosage formula for phased genotypes (x[1] + x[3]) assumes the same column ordering as bgen_reader — if bgen's variant.probabilities arranges phased probabilities differently, the dosage would be silently wrong. If your test data was unphased, it would be good to confirm the phased path is correct before we merge.