Deduplicate closures

[oli-obk]

#62429 manually deduplicated monomorphizations of closures. We should automate this during building instead of requiring users to do it manually. Ideally we'll revert the parts of #62429 that made things less readable as the last step of closing this issue.

I think the endgame would be that we'd deduplicate closures (in release mode only, so backtraces's file:line locations still make sense in debug mode), as long as their body is the same.

Not sure if that is possible, but random idea:

If we only care about monomorphizations in llvm, we could create a function that takes a closure's mir::Body, cleans it of all "unnecessary" info like Spans and then feeds that through a query (which thus will only be called once with the same arguments), producing the appropriate codegen backend's function handle.

If we want to reduce type lengths, too, we'd need to change the way closure substs are built, by computing the minimal required set ahead of time.

[cuviper]

Deduplicating is nice, but I believe LLVM is already pretty good about this, so I don't expect too much benefit there. The exploding type lengths were the real practical thing that #62429 addressed. I probably should have emphasized that better in the description.

[Centril]

@eddyb in #62429 (comment):

Isn't this #46477? We should revive those efforts and implement it as "polymorphic" codegen (still-generic over some parameters which aren't actually used), in a similar way to how "polymorphic" CTFE works with miri today.

[jonas-schievink]

Seems like a clear duplicate, so closing in favor of that.