Usage of global scope by Linker is unsafe for pointers to `static` data in loaded library

[some-java-user-99206970363698485155 at vodafonemail.de]

Thanks a lot for looking into this and the detailed suggestions!

> When a client passes an arena to a library, it is effectively 
> *transferring ownership* of that arena to the library
I guess technically it is more "sharing" ownership here, because the 
user should still be able to use the Arena as well (e.g. to unload the 
parser library eventually by closing the Arena), and `Language#load` 
respectively the created Language object should not and will not handle 
closing of the Arena I think.


> In general, the idiom where you have a method that receives a random 
> segment and you do a reinterpret on it, attaching a cleanup action 
> also looks suspicious. The method doesn't have "ownership" of the 
> segment, so how do we make sure we're not really adding _two_ cleanup 
> actions for the same segment?
That is a good point. I think the `Language(MemorySegment)` constructor 
currently attaches the `ts_language_delete` cleanup because that 
function is defined in the tree-sitter general library, not in the 
specific parser. But you are right, I think if you created multiple 
Language objects from the same language MemorySegment, that would lead 
then to a double-free because `ts_language_delete` is called multiple 
times. Luckily this won't happen at the moment because 
`ts_language_delete` seems to be no-op [1], see 
https://github.com/tree-sitter/tree-sitter/blob/v0.23.0/lib/src/language.c#L13-L17.


Kind regards