Allow shots to be a numpy.int64

[garrison]

Summary

I've noticed that the Aer backends all work properly if the shots count is passed as a numpy.int64, but the BasicAer and hardware backends both fail in this case. The simplest way to make it work on all backends is by making this change in terra, inspired by #4591.

Details and comments

Basic test program for reproducing the bug:

import numpy as np

from qiskit import QuantumCircuit, execute, Aer, BasicAer, IBMQ
from qiskit.providers.ibmq import least_busy

qc = QuantumCircuit(2, 2)
qc.h(0)
qc.cx(0, 1)
qc.measure([0, 1], [0, 1])

def run(backend):
    sim = execute(qc, backend, shots=np.int64(1024))
    return sim.result().get_counts(qc)

# This works
run(Aer.get_backend("qasm_simulator"))

# This fails, but is fixed by the current change
run(BasicAer.get_backend("qasm_simulator"))

# This also fails, and is also fixed by the current change
provider = IBMQ.load_account()
run(
    least_busy(
        provider.backends(filters=lambda x: x.configuration().n_qubits >= 2, simulator=False)
    )
)

[coveralls]

Pull Request Test Coverage Report for Build 2161409796

• 0 of 0 changed or added relevant lines in 0 files are covered.
• 6 unchanged lines in 1 file lost coverage.
• Overall coverage increased (+0.0002%) to 83.944%

---

Files with Coverage Reduction	New Missed Lines	%
qiskit/compiler/assembler.py	6	96.23%

Totals
Change from base Build 2157761883:	0.0002%
Covered Lines:	54202
Relevant Lines:	64569

---

💛 - Coveralls

[mtreinish]

I think it would be better to use numpy.integer here instead of the abstract numbers class from stdlib. We've tried to use the abstract numbers classes before and had performance regressions each time (because of how the abstract classes hook into the checks for isinstance). While assemble isn't super performance critical it's still used in job submission path a surprising number of place and this is only just a single isinstance call so the extra overhead isn't as huge as something checked on each instruction in a circuit, I still think it would probably be better to just use numpy.integer even if that means other int types from other external libraries can't be used. In practice I don't think anything besides numpy ints will really come up.

[garrison]

Sounds reasonable; I've pushed a fix.