Fix global phase tracking for Unroller, OneQubitEulerDecomposer, and ControlledGate

[georgios-ts]

Summary

Related and (partially) resolves #5235. Fixes some bugs found in handling global phases.

Details and comments

The PR introduces the following changes:

• Decomposer and Unroller may add twice a global phase if a gate has more than one instructions. See below. This has been fixed.

qc = QuantumCircuit(1, name='V', 
                    global_phase=pi / 2)
qc.x(0)
qc.h(0)
V = qc.to_gate()

circ = QuantumCircuit(1)
circ.append(V, [0])

print(circ.decompose().global_phase)

pass_ = Unroller(basis=['x', 'h', 'cx'])
pm = PassManager(pass_)
print(pm.run(circ).global_phase)
---
Outputs:
3.14
3.14

• Tracks global phase in OneQubitEulerDecomposer. In this way, the global phase in the definition of a 2q- UnitaryGate is stored. Similar with Fix decomposition in UnitaryGate.definition for 1Q gates #4996 . Previously,

pauli = Operator(Pauli(label='XZ'))
gate = UnitaryGate(pauli)

np.allclose(pauli.data, 
            Operator(gate.definition).data)
---
False

• Creating a controlled version of a gate, a phase may appear after _unroll_gate (L115). This wasn't previously tracked and it was the cause for Incorrect relative phase added for controlled Pauli-X from quantum_info.operators #5235 .