[Feature Request] Add support for evaluating baseline

[m-roberts]

Currently, only new experiments can be created. However, in some cases, such as running backtests on a new version of an agent, it is necessary to run evals on an experiment that doesn't have evaluation results.

By adding support for evaluations to be run on a specific example, evaluations can be run against a baseline experiment - useful when converting production traces to an experiment.

This can be achieved by allowing an experiment name to be provided in a config file instead of a dataset, target function, experiment prefix, etc.