"_mac" might mean "macromodel" which is a common approach
to getting RF simulation fidelity (embodying more of the close-in
parasitics that the compact model has nothing, for).
That's just one conjecture. The two (nch_lvt, nch_lvt_mac)
may have the same core layers and geometry but perhaps
the PDK has some "recognition feature" (like, say, "RF/drawing"
layer) that distinguishes the "just digital" device from the
"RF" device. Perhaps this is just bookkeeping, or perhaps
there are -supposed to be- specific features of the "RF"
device (such as gates strapped with Met1 and contacted
at both ends, a fixed or constrained W, etc.) and so an
enforcement of "tell me the same story the same way
twice" discipline makes more sense (preventing a surprise
of some sort).
Or maybe nch_lvt_mac is the device you're supposed
to use, and nch_lvt exists only to map the core device
of the macro to its model and has no layout view of
its own.
You could investigate these conjectures in the PDK
structure if you didn't have the more useful task of
fixing your schedule-bog by way of achieving design
consistency.
I'd bet there is a lovingly crafted library / modeling
document somewhere in said PDK that tells what's
what, feeling all lonesome because nobody ever
reads it.