Caching grouped results and refactor sampling estimator

[dchung0741]

New changes

• Caching mechanism: The CachedMeasurementFactory class decorator is supported to cache the outputs of measurement factories.
• The function body of (concurrent_)sampling_estimate is refactored while keeping the interface intact.
  • helper functions provided:
    • distribute_shots_among_pauli_sets: distribute shots among distinct commuting sets.
    • CircuitShotPairPreparationFunction: type alias that represents a function that prepares the (circuit, shot) pairs for performing sampling estimation on a state.
    • get_sampling_circuits_and_shots: default CircuitShotPairPreparationFunction that concatenates measurement circuit after state preparation circuits.
    • get_estimate_from_sampling_result: returns the estimate from sampling count.
  • Slight change in the interface (not breaking): New circuit_shot_pair_prep_fn argument is added to (concurrent_)sampling_estimate. It is default to the get_sampling_circuits_and_shots helper function, so old codes are still supported

Examples

• How caching is supported

cached_measurement_factory = CachedMeasurementFactory(bitwise_commuting_pauli_measurement)
estimate = sampling_estimate(
    Operator({pauli_label("Z0 Z7"): 1, pauli_label("X1 Y2"): 8}),
    quantum_state(8, bits=1),
    1000,
    concurrent_sampler,
    cached_measurement_factory,
    allocator,
)

[github-actions]

🚀 Deployed on https://deploy-preview-302--quri-parts.netlify.app

[rykojima]

Thank you for the implementation. LGTM!