After the previous pr some problems were noticed, this pr should solve them.

This appears to effectively be a partial revert of #1, which re-introduces nvaccess/nvda#16027 (closed as "intended behaviour" in upstream per JoshData/fast_diff_match_patch#26 (comment)). I think the difflib fallback is probably the best we can do under the circumstances unfortunately.

Are you able to reproduce the issues noted in nvaccess/nvda#16483 (comment) by using utf-32 instead of utf-8?

Please read conversation at nvaccess/nvda#16483

This revert will not lead to problems with emoji processing, since this problem is solved on the fast diff match patch side.

With the current implementation, I'm seeing significant regression, for example when restarting Home assistant using ha core restart, a large number of exceptions are thrown, probably about 10 per second.

fast diff match patch simply not working with utf-32.
I'll double check now just in case.

This revert will not lead to problems with emoji processing, since this problem is solved on the fast diff match patch side.

Ah nice!