Support acquisition-function-speciic tensorboard logging

[uri-granta]

Describe the feature you'd like
At the moment, the default multi-objective pairplots can't take e.g. constraints into account: all they know is that there are multiple tagged datasets, not what those datasets represent.

It would be useful if we could generate tensorboard plots that are specific to an acquisition function builder: e.g ExpectedConstrainedHypervolumeImprovement does know both the data and the constraints. We already do something similar for models, with their log method.