[llvm] [mlir] [mlir] split transform interfaces into a separate library (PR #85221)

[Stella Laurenzo via llvm-commits]

stellaraccident wrote:

> I am curious to learn more about cases that worked for you and those that did not. There are intentionally no recommended use cases so as to not limit what people can do with it but rather learn from the emerging use cases. Unfortunately, I haven't thought through the information gathering strategy, so most of the feedback I heard so far boils down "it didn't work for me" without any other actionable information. There seems to be a common problem of people conflating the idea of capturing (parts of) IR transformation in the IR itself with the limited set of transform ops currently available upstream.

I'll be at eurollvm and might be a good discussion to have over drinks. It definitely produces some polarizing outcomes (I think I'm past opinions at this point and more judging based on what worked and didn't over many months/years) that make me see a couple of sides.

https://github.com/llvm/llvm-project/pull/85221