Pheweb

- Speeds up `pheweb gather-pvalues-for-each-gene` ~2x by avoiding reading any variant twice. (Thanks to finngen for this suggestion.)
- Allows live-filtering a manhattan plot by MAF or snp/indel, with instructions in README.
- Verifies that `num_cases + num_controls == num_samples` in `pheweb phenolist verify` (which is included in `pheweb process`).

- Removes dependence on `pandas` (because it wouldn't install on my laptop)

- Allows setting `loading_nice = True`.
- Allows setting `field_aliases` again.
- Reduces memory usage by `pheweb qq` by ~10x by switching to `numpy` and `pandas`.
- Fixes the bug where `pheweb matrix` breaks when `matrix.tsv.gz` is up-to-date.

- Rewrites configuration management, losing the ability to customize `extra_per_*_fields` and `null_values` and `field_aliases`.
- Fixes bug where config wasn't passed to child processes when using `PHEWEB_DATADIR` or `pheweb conf key=value <subcommand>`.

Bugs:

- `pheweb matrix` breaks when `matrix.tsv.gz` is already up-to-date.

- Makes sure that `pheno_gz/<phenocode>.gz.tbi` gets created, and re-runs traits that don't have it.

- Uses dbSNP v154 (the latest!) with way more rsids. To use them, run `rm generated-by-pheweb/sites/sites-rsids.tsv && pheweb process`.