Move unitary matrix circuit simulator to qiskit-synthesis

[jakelishman]

This isn't the most natural place ever for it, but it wasn't really any better in qiskit-quantum-info, and this doesn't produce problematic ordering in the dependency chain.

Depends on #15912, but only in the sense that one of the two moves needs to set up the qiskit_synthesis::matrix module, and I did that one first.

[qiskit-bot]

One or more of the following people are relevant to this code:

• @Qiskit/terra-core
• @levbishop

[ShellyGarion]

I think that originally putting unitary_sim in the quantum_info crate was done to mimic the python code structure where all simulators (unitary, statevector, clifford, stabiliser) are in quantum_info.
I think that eventually if we'll have more than one simulator we should have a seperate crate for them.
But for now putting it in the synthesis crate makes sense as it's only used in mcx.rs

[jakelishman]

Shelly: yeah, that's totally fair for why it was there - there was also a bunch of code in qiskit-circuit that was there not because it was actually related to the circuit IR, but because that just happened to be our "lowest" crate before now.

We actually have this same dependency problem within Python too, but we use lazy / scoped import statements to get around it there. We don't have that same option with separate Rust crates (and it's not a good situation in Python anyway), so this is necessary to get the dependency graph we want.

[ShellyGarion]

LGTM, thanks!