Split handling of standard conforming DICOM from non-standard DICOM

As suggested by @pieper in https://github.com/InsightSoftwareConsortium/ITK/pull/4134#issuecomment-1691734111

> My general preference would be to have the core reader code only handle completely standard dicom so that it is readable and maintainable. Any special code to handle malformed or vendor specific data should be detected and handled in dedicated classes, ideally with diagnostic info available so users know they have odd data and can report the issues to whoever created it.