Resume experiments fail when trying to bulk register previous metrics to the optimizer

[eujing]

During bulk register, existing metrics from the results db are loaded as strings as that is how they are stored.

However, applying self._adjust_signs_df to the string target columns results in un-intended results, especially when in a max case.
The string column *= -1 results in empty strings, which fail to cast into float later on in bulk register.
This issue probably never turned up in min cases as the string columns *= 1 has no effect.

The casting to float should probably be done on all target columns before adjusting signs.

Example output from the debug terminal: