Is it an error in tools or the XML is really not well-formed?
It's well formed in the latest specification, which is XML 1.0 Fifth Edition. But it was not well-formed in the previous edition, which was current until 2008.
The original XML 1.0 spec (from 1998) locked down the set of name characters to the characters that were defined as letters in the Unicode standard of the time. That didn't include
which only came in with Unicode 3.1 a few years later.
XML 1.1 was much looser about what characters it would accept in names (largely for this reason, to allow characters from future Unicode versions), and this is a Good Thing. However XML 1.1 has never really caught on, so the Editors decided to backport the newer, more permissive namechar rules from there to 1.0. This was controversial and all in all probably not a Good Thing.
This means you can use
in names in XML 1.0 documents and be usable by a subset of parsers that have updated for Fifth Edition (or never implemented the strict rules in the first place), or you can use them in XML 1.1 documents and be usable by a different set of parsers that support XML 1.1.
Or, more realistically, you can avoid those characters which are sort-of-well-formed-depending altogether, and feel a little sad.