VCF field conversion bugs

[tomwhite]

Fixes #353 and #355.

However, there is another failing case (like #353), shown by the INFO/HWE_EUR_unrel field in 1kg_2020_chr20_annotations.bcf, which has Number=A, but max_number=0. To fix that, requires finding the actual size of A (alt alleles) for the whole dataset.

There's a comment about this:

bio2zarr/bio2zarr/vcz.py

Lines 143 to 144 in 17db733

	# TODO we should really be checking this to see if the named dimensions
	# are actually correct.

@jeromekelleher have you thought about how to implement this?

[jeromekelleher]

I'm happy with this as way of squashing these bugs, but I think we'll probably need to be a bit more systematic about dealing with the dimensions to cover all the bases. I guess we'll just need to have a special code path for each of the named dimensions so that they can be handled correctly, taking into account that we could easily have incorrectly stated dimension sizes (i.e., something that says it's allele dimensioned, but is actually of size 10 or something).

You're going to need to update the stupid du tests I'm afraid...

[coveralls]

coverage: 98.414% (+0.003%) from 98.411%
when pulling 0b0d8e5 on tomwhite:vcf-field-conversion-bugs
into 17db733 on sgkit-dev:main.

[tomwhite]

I opened #359 for the case that is still failing.