Kill CSEs. Add substitution rules, precomputes.

loopy/__init__.py

loopy/codegen/loop.py

loopy/cse.py

loopy/kernel.py

loopy/preprocess.py

loopy/schedule.py

loopy/subst.py

loopy/symbolic.py

test/test_fem_assembly.py

test/test_linalg.py

test/test_loopy.py