For Illumina reads, RG typically groups reads from a lane. GATK requires/assumes such a use in several components. At least one should not put reads from multiple libraries in the same RG.
When reading the input BAMs, samtools tries to group reads based on their @RG-SM tag. If there are no @RG lines, it implicitly inserts the following line into the header:
@RG ID:filename SM:filename
The VCF spec is very clear about the definition of QUAL: the probability of there being no SNPs in any samples.
You can't always consider @RG the same as sample name. For example, when we do high-coverage genomes, we often have to create several libraries to get enough DNA, and each of those libraries may be sequenced across multiple runs. At the end of the process, we want to lump all of these reads together to make variant calls. The @RG tag allows us to combine all of these reads (from the same sample) into a single bam, while still retaining information on the source of each read.
As drio says, there's an SM tag within the field that corresponds to sample name.
In multi-sample VCFs, you shouldn't be using the QUAL field for the quality of the SNP calls. You should define a FORMAT field to store it - I've often seen VAQ for "variant quality". That way, each sample has an independent score at each site.
@RG allows you to group together a set of reads within a (s|b)am file. In the @RG you have the SM tag to specify sample information.
Yes. Imagine you are working in a project and need a lot of sequence coverage. You typically want to create more than one library to avoid PCR duplication issues.
The LB tag in @RG for each library would be different. Once you merge your alignments, you'll see in the header multiple @RG entries.
