improve performance of tree sitter query captures (for text object motions in particular)

[pascalkuthe]

Helix sometimes freezes when performing treesitter motions on medium to large files.
The problem only occurs in very particular cases and depends highly on the grammar, specific capture and file content.
I was able to reproduce the problem consistently by using any treesitter motion in the iter.rs file of ropey.
But @dead10ck also noted that the same problem sometimes occurs for him in the matrix chat.

Looking at the flamegraph almost all time is spent in treesitter itself:

After a bit of digging I found the problem:
Since treesitter version 19.5 treesitter does not set any limit for the amount of in-progress captures any more (used to be 32).
Because advancing the QueryCursor requires comparing every new capture to all in-progress captures, this can become O(N^2) in pathalogical cases and freeze the editor.
The issue is solved by setting a match limit.
I discovered this fix while browsing the neovim treesitter code (as it does not suffer from this problem) where a limit was introduced after the 19.5 release
to fix similar issues. I have added more details (and pointers to neovim) in comments.

To avoid similar problems I also made sure this limit is set in all other places a QueryCursor is created (in the syntax highlighting).
I tried out a few large files where I had performance problems with tree sitter highlighting in the past, but this fix doesn't help much there.
However, neovim did introduce this limit to fix performance issues with syntax highlighting, so I think it's good to add the limit to syntax highlighting just in case.

While investigating this problem I found a few other ways to slightly improve the performance of the treesitter movements.
I bundled these here as they are too small to really warrant a seperate PR.

[pascalkuthe]

The tests failed on the last PR run but they pass locally it seem spurious to me (some kind of permission errors).
I had to push some fixes to the docs anyways so I hope its fixed now.

[pascalkuthe]

Ok nvm I actually missed the real errors from the permission errors. Those permission are unrelated but way more common then the test failures. The core change is not the problem only the small I made to the treesitter movement. I just dropped those changes for now and will open a followup PR if I get them fixed