perf: speed up LP file writing (2.5-3.9x on large models, no regressions on small)

perf: speed up LP file writing (2.5-3.9x on large models, no regressions on small) #902