create modules for jaxqtl

[zixuanzhang]

modules including:

• cis/nominal scan for eQTL mapping
• regression of gene expression on non-genotype covariates
• logistic regression of PIP for cis-SNPs per gene on annotation file (i.e., nominal scan on each col of annotation)

[zixuanzhang]

For running logistic regression on PIP, jaxQTL gives different beta and SE estimate (but similar p values) compared to statsmodel result. This is caused by different stopping rule. jaxQTL used diff in log likelihood whereas sm used diff in parameter estimate (jaxqtl can replicate sm results if use the same stopping rule).
In this case running logit on PIP instead of 0/1 outcome, using Bernoulli likelihood is probably suboptimal since here y is not Bernoulli random variable.