Rework validation for measurements and pending_experiments

[Scienfitz]

Fixes #453

Notes re measurements validation

• Two utilities (one for parameters and one for targets) for input validation has been extracted. Additional validations for binary targets have been added. The utilities contain parts from add_measurements and fuzzy_row_match
• As a result fuzzy_row_match does not perform any validation anymore
• add_measurements now simply calls the utilities
• tests for invalid parameter input have been extended

Notes re pending_experiments validation

• The parameter input validation utility is now called as part of Bayesian recommenders, causing proper error messages instead of crypto and seemingly unrelated ones
• A recommender based test for invalid values in pending_experiments has been added

Remaining problem:
The places of validation for measurements and pending_experiments are inconsistent. The former is done in the campaign and not in any recommender, the latter is not done in the campaign but done in the recommender. One of the issues is that numerical_measurements_must_be_within_tolerance is not available for recommenders, but its required for the parameter input validation.

Suggestion:
Imo we dont want to add more keywords to .recommend or so, hence the pending_experiments validation currently assumes numerical_measurements_must_be_within_tolerance=False and cannot be configured otherwise. To me it seems the simplest solution would be to completely get rid of any numerical_measurements_must_be_within_tolerance keywords and make it an environment variable, we wouldn't have to worry about its availability anymore and it would require adding it to more and more signatures.

[AdrianSosic]

Hi @Scienfitz, thanks the refactor 🏗️ Below my comments

[AVHopp]

LGTM