This way, we won't store each welcoming message in the database, but
updating the contstant message ID once the mesage changes allows us to
still know exactly what the user was shown.

Icons from https://feathericons.com/

now that each message has its own id and feedback is related to this, we
can allow feedback for any previous message.

Because we do post-processing of the LLM response (e.g. replacing tags
with links) it is difficult to handle streaming, so we give it up
completely.

It was getting unwieldy to modify;
also the combination of html+css+js meant it was hard to edit it
(syntax highlighting, comments, autocompletion etc.)