Update all emulator's returning independent distributions to use the `Independent` distribution

[sgreenbury]

Following #924, since there are cases where probabilistic emulators return a distribution that has is not Independent (e.g. GaussianLike(..., DiagLinearOperator(...)), we should aim to update these cases if possible to ensure there are no cases that are not supported by metrics.

Ideally we could encode this with subclasses similar to proposed in #436 but it might be sufficient to replace these cases with Independent(Normal(...), ...) but we should investigate.