Make quadruple precision the default

[mdickinson]

It would be nice to make the default context correspond to IEEE 754 binary128 format. That's a more immediately useful default than double precision, since we already have that in the form of the native float.