Support weighting samples before loss aggregation

[han-ol]

It would be useful to handle sample weights in all loss computations.
sample_weights can be an additional keyword argument for compute metrics akin to inference_variables and inference_conditions.

Users would specify sample_weights in their simulator and configure an adapter accordingly:

adapter = bf.ContinuousApproximator.build_adapter(
    inference_variables=["x"],
    inference_conditions=["y"],
    sample_weights=["sample_weight"],
)