(See pyston/python-macrobenchmarks#10.)

• a benchmark group "all" is allowed, but ignored
• duplicate groups are allowed (between manifest files) but the semantics aren't clear
• a group will overwrite a tag with the same name, without warning
• the "list_groups" command does not show tags if any custom groups are defined
• tags aren't found properly in BenchmarksManifest.resolve_group()