Avoid superfluous copy of residual vector when applying preconditioner

The residual vector is changed from a jax-numpy array to a standard
numpy array before the preconditioner is applied. Now numpy is told
to avoid copying the residual data and instead re-use the buffer.
This should avoid overhead, especially for large vectors.

Author: btalamini

optimism/SparseCholesky.py

@@ -51,7 +51,7 @@ def update(self, new_stiffness_func):
 
     def apply(self, b):
         if type(b) == type(np.array([])):
-            b = onp.array(b)
+            b = onp.array(b, copy=False)
         return self.Precond(b)