FullPeepholeOptimise targeting CX instead of TK2

[sjdilkes]

We have found some cases where synthesis to ZZPhase via TK2 is sub-optimal. Fixing the pass is the main priority, but we may be able to achieve the same performance by just targeting CX in optimisation level 2 and then rebasing.

[yao-cqc]

We should be careful not to bring back Quantinuum/tket#1738.