[llvm] [LLVMABI] Implement the ABI Typesystem (PR #158329)

[John McCall via llvm-commits]

rjmccall wrote:

`.6%` was the best number we had. You are right that it wasn't statistically significant, but that just means it was weak evidence, not that it should be dismissed; a 50% increase on one sample also wouldn't be statistically significant, but it's obviously evidence of a problem.

Anyway, your new numbers are good evidence that this isn't a problem. Thank you for running them.

https://github.com/llvm/llvm-project/pull/158329