Add `B` gate to standard gate set, and teach compiler about decomposition in it

[stylewarning]

Add the B gate, which is known to lead to optimal decompositions of 2Q gates into only two B gates. That is, let G be an operator in PU(4). Then

G = L1 * B * L2 * B * L3

where Ln are one-qubit gates.

The definition in the paper is (verbatim):

B = exp(i/2 * pi/2 * XX) exp(i/2 * pi/4 * YY)

In addition to defining B, write a few define-compilers for it, especially with respect to other 2Q gates.

Note: This would change the Quil spec.