overhaul mapping methodology for lsm/resource/modelnet
Currently, the APIs used for sending messages to (or through) the LPs depend on an implicit mapping to LP-IDs in the same configuration group as the caller. Where configuration "offsets" are concerned, a hack is performed that simply modulo's the caller offset against the number of callee offsets. For trivial 1-1 or N-1 mappings this is OK but anything more and this is insufficient. Case in point, connecting different lp types to dragonfly terminal LPs.
Need to think of a way to provide better mapping flexibility, preferably without completely overhauling how we do mappings.