Improve documentation on `cnvkit call` and the use of VCFs

[tskir]

Currently, some aspects of how VCF samples are loaded and processed are a bit obscure in the documentation. This especially applies to things like BAF rescaling and LOH detection; see for example https://github.com/etal/cnvkit/issues/601#issuecomment-835861846.

It seems like a good idea to restructure and clarify the parts related to VCF, how the samples are treated, how the maths of BAF rescaling and LOH detection works, etc.