TargetSchedModel now always delegates to MCSchedModel the computation of instruction latency and reciprocal throughput. No functional change intended. llvm-svn: 330099