Support more characters in feature names

This dataset currently can not be processed. As far I can tell it is because in this version a city is represented as nominal (instead of string). This seems like an issue for the feature description xml. The particular violation it finds (first) is Bayamón (and ó is not basic latin). I presume some (or all) of this will be reworked when moving to parquet (see #388), but I figured I would document it is currently an issue.

Problem validating uploaded description file: XML does not correspond to XSD schema. Error Element '{http://openml.org/openml}nominal_value': [facet 'pattern'] The value 'Bayamón' is not accepted by the pattern '\p{IsBasicLatin}*'. on line 170 column 0. Error Element '{http://openml.org/openml}nominal_value': 'Bayamón' is not a valid value of the atomic type '{http://openml.org/openml}basic_latin256'. on line 170 column 0. 

Reproduce the upload validation failure with dataset 42725 and inspect the XML/XSD rule named basic_latin256, especially the nominal_value pattern. Determine the compatibility implications noted in #388, then verify that feature names such as Bayamón are accepted without breaking the existing description validation.