It turns out that Sentry has various contraints on the characters that
can appear in tags as well as their length. For debugging purposes we
want to capture the full document URL unmodified. Therefore capture this
as un-indexed "extra" metadata, which has fewer constraints, rather than
an indexed "tag".