LLM-driven optimizations

[David Neil]

I gave an LLM a couple of days with the goal of making our internal closure-compiler pipeline faster.  The end result was getting wall time down from 290s to 100s for a byte-equivalent version, and down to 60s for a less optimized version that is 0.5% bigger.

Here are the changes: https://github.com/lucidsoftware/closure-compiler/pull/3/changes

I am curious if there is any appetite for these changes being merged into the master branch?
We are still reviewing the changes internally, but the byte-equivalent output is promising.  We don't use Java ourselves, so there isn't a lot of expertise around to review these changes internally, so if they are going to be merged upstream I would like to skip adding them to a fork so that I don't have to reconcile the two branches later on.

If there is interest I can open a PR, I just didn't want to thrust an 8000 line diff on someone without interest from the maintainers first.

Thanks.

--David