Fix density matrix simulator when subcircuits exist

[daxfohl]

Fixes #3723 by forcing run_sweep_repeat in density matrix simulator if subcircuits are detected in the circuit. run_sweep_sample does not work in this case because there's no way to sample a subcircuit.

[95-martin-orion]

This is reasonable as a workaround for #3723, but run_sweep_repeat takes O(repetitions) times longer to complete than run_sweep_sample. In the long term, we'll want to switch back to "sample" where possible. The same applies to sparse_simulator, which will mark an entire subcircuit as non-unitary even if it only has terminal measurements.

One cheap-ish way to do this would be to split the subcircuit and its measurements:

# pseudocode
Circuit(CircuitOp(op1, op2, op3, measure))
--> Circuit(CircuitOp(op1, op2, op3), measure)

That step needs to be recursive, but it would be much cheaper to implement "subcircuit splitting" than full subcircuit sampling.