+5. Profile the construction of "large" matrix algebras (like the
+ 15-dimensional QuaternionHermitianAlgebra(3)) to find out why
+ they're so slow.
+
+6. We should compute whether or not the algebra is associative if it
+ is unknown. I guess the "associative" argument should be ternary
+ (True, False, None)? We should also figure out the correct
+ True/False values for the example classes, and of course add an
+ _is_associative() method.
+
+7. When field=RDF, subalgebra construction is failing because the
+ inner product isn't associative? Actually, it's the combination
+ of field=RDF and orthonormalize=True.