update cpu pipeline

- do not bufferize func boudaries, we already use memrefs
- add canonicalize before remove-temporaries
- add deallocation pass

Author: tkarna

src/jit/mlir.cpp

@@ -688,8 +688,10 @@ static const std::string cpu_pipeline =
     "arith-expand,"
     "memref-expand,"
     "func.func(empty-tensor-to-alloc-tensor),"
-    "one-shot-bufferize{bufferize-function-boundaries},"
+    "one-shot-bufferize,"
+    "canonicalize,"
     "imex-remove-temporaries,"
+    "func.func(buffer-deallocation),"
     "func.func(convert-linalg-to-parallel-loops),"
     "func.func(scf-parallel-loop-fusion),"
     "drop-regions,"