Speed up assembling of noise model for circuits without delays

[itoko]

Summary

The commit fixes the remaining minor performance issue introduced at #1391 by skipping the run of noise pass manager for circuits without delays.

Fixes #1407 (performance regression) completely in addition to the major fix #1408.

[itoko]

I've confirmed that this PR significantly reduces the assemble time of noisy simulation of circuits without delays.

Without this PR:
TRANSPILE TIME TAKEN: 66.19
ASSEMBLE TIME TAKEN: 5.05
RUN QOBJ TIME TAKEN: 6.21

With this PR:
TRANSPILE TIME TAKEN: 63.55
ASSEMBLE TIME TAKEN: 1.17
RUN QOBJ TIME TAKEN: 6.41

Using the following code

import time
import numpy as np
from qiskit_experiments.library import StandardRB
from qiskit.providers.aer import AerSimulator
from qiskit.test.mock import FakeParis

backend = AerSimulator.from_backend(FakeParis())

start = time.time()
lengths = np.arange(1, 1000, 100)
num_samples = 10
seed = 1010
qubits = [0]

# Run an RB experiment on qubit 0
exp = StandardRB(qubits, lengths, num_samples=num_samples, seed=seed)

# Step by step
from qiskit import transpile

# Generate and transpile circuits
start = time.time()
circuits = transpile(exp.circuits(), backend)
stop = time.time()
print("TRANSPILE TIME TAKEN: {:.2f}".format(stop - start))

# Simulator Assemble
start = time.time()
qobj = backend._assemble(circuits)
stop = time.time()
print("ASSEMBLE TIME TAKEN: {:.2f}".format(stop - start))

# Simulator Run
start = time.time()
result = backend._run(qobj)
stop = time.time()
print("RUN QOBJ TIME TAKEN: {:.2f}".format(stop - start))

[jakelishman]

This looks fine to me, but #1409 also has the same optimisation, but with less data iteration due to a new data structure it introduces. I think #1409 likely supersedes this.

[chriseclectic]

As @jakelishman said this is already covered by #1409 which includes this fix along with a fixes for other unnecessary iteration from qasm3 compiling, qerror_loc insertion, and qobj splitting.

[itoko]

Close as it's already covered by #1409