The work done to add typing to datagrids regressed the storage of

datagrid columns. We were converting a comma-separated string of column

names to a list of strings for processing, but then later comparing that

to the stored comma-separated string.
This provided a mismatch and caused us to override the stored setting

with a serialized version of that list. However, that was a Python

representation of a list of strings, and not a CSV. On next load, that

was considered invalid, and we'd fall back on the datagrid's defaults.
The regression was never released, but encountered on

https://reviews.reviewboard.org.
This change fixes this, retaining colnames as a CSV and instead using

a separate variable for the normalized list for processing. It also

fixes a typo in the typed profile_columns_field name, and adds unit

tests.

Unit tests pass.
Manually tested column customization and sorting, verifying that my

changes persisted.