We would occasionally hit errors when loading datagrids, typically

caused by web crawlers. In the case where the user profile or the

querystring specified column names, but none of them were valid, we'd

progress through building the (empty) list of columns and then crash

trying to index into it.
This change moves things around a bit so that instead of just testing to

make sure the comma-separated list of column names isn't empty, we try

to load all of the columns and then check if that resulting list is

empty, defaulting to the standard set of columns if so.

Loaded /users/X/&columns=,. Before this change, it would crash with

an IndexError. After, it loads successfully.