People do cleanroom implementations as a precaution against a lawsuit, but it's not a necessary element.
In fact, even if some parts are similar, it's still not a clear-cut case - the defendant can very well argue that the usage was 1. transformative 2. insubstantial to the entirety of work.
"The complaint is the maintainers violated the terms of LGPL, that they must prove no derivation from the original code to legally claim this is a legal new version without the LGPL license."
The burden of proof is on the accuser.
"I am genuinely asking (I’m not a license expert) if a valid clean room rewrite is possible, because at a minimum you would need a spec describing all behavior, which ses to require ample exposure to the original to be sufficiently precise."
Linux would be illegal if so (they had knowledge of Unix before), and many GNU tools are libre API-compatible reimplementations of previous Unix utilities :)
Now, whether chardet 7.0.0 is a derivative of chardet or not is a matter of copyright law that the LGPL has no say on, and a rather murky ground with not that much case law to rely on behind it. If it's not, the new author is free to distribute chardet 7.0.0 under any license they want, since it is a new work under his copyright.
The original code is part of claude's training material. With that intepretation of the LGPL AI is incapable of writing non LGPL derivatives. I like that interpretation.