To my inverted way of thinking there is something wrong with the general
logic of what is being attempted here.
I think people should be encouraged to upgrade to the most recent
released version of R. At best I test my packages with the last two
releases of R, so it is only by accident that the "Depends" line on my
packages may not specify the most recent release.
On the other hand, an unmaintained package could quickly depend on an
older version of R. I don't see any way a package maintainer can specify
this, but there could be some mechanism (possible tied to passing R CMD
check) that specifies dependency on an older version of R. Perhaps this
might also be formally recognized as a mechanism for moving things out
of the contrib section of CRAN. If this is to be written into a file in
the package it would require modifying the contents of the tar ball, but
if it could be coded into a prefix on the package name it might be
simpler?