Better TK2 Squashing

[sjdilkes]

Related to #971.

SynthesiseTK doesn't recognise XXPhase(a,0,1).YYPhase(b,0,1).ZZPhase(c,0,1) == TK2(a,b,c,0,1) and running FullPeepholeOptimise on YYPhase(b,0,1) returns a TK2(b,0,0,0,1) with a bunch of TK1 single qubit gate wrappers, rather than TK2(0,b,0,0,1).

This isn't too disastrous in most circumstances, as if the circuits are submitted with ZZPhase then this would happen anyway. However, if native TK2 gates are being used then these amount to extraneous single qubit gates. Additionally, the original reason I had this come up was for compiling circuits to the Iceberg code. In that case, encoded TK1 are as expensive (~) as encoded TK2, so pre-compiling with FullPeepholeOptimise and then encoding led to large, extraneous gate overhead.

My guess is that if the code here can recognise these improvements then everything else should work: https://github.com/CQCL/tket/blob/a443386f8c3703e6a1afb997d9f44460e59fc3d6/tket/src/Transformations/BasicOptimisation.cpp#L143C3-L152C4

Hopefully that doesn't have to come to improving two_qubit_canonical.