Printer: add support to print types in MLIR style

[tobiasgrosser]

This brings the xDSL output closer to the syntax MLIR is using.

[webmiche]

I think this is a great idea, but I would like to see some test cases. Afterward, we can merge 👍

[math-fehr]

Thanks for this patch, I just added two nits comments.
I think I made a bad choice by having types closely linked with the values. While I think this makes more sense that the MLIR way, I believe that the change is not good enough to differ from MLIR.
I think this option should be the one by default, though we should give time to people to update their project with this version, before changing everything!

[Dinistro]

Maybe add a test or two, otherwise LGTM.

[math-fehr]

Actually, my bad, I want to add more things to it before merging it. I'm not sure how to make this a draft PR though.
I would like to essentially print correct MLIR (without going through the converter) as well.

[webmiche]

Actually, my bad, I want to add more things to it before merging it. I'm not sure how to make this a draft PR though. I would like to essentially print correct MLIR (without going through the converter) as well.

You can make it a draft in the reviewers section of the PR (not a very natural place to be IMO ,but it is :))

[math-fehr]

Okay, that should now be ready for review, I made quite some changes since the last review!
The PR adds an MLIR target in xDSL, which allows to print MLIR instead of xDSL. This should allow to directly print an MLIR program without having to install MLIR. I made this the default of the -t MLIR target in xdsl_opt_main, @webmiche tell me what you think about this.
Attribute can override a method to have a custom printing format for MLIR (I'll add a similar thing for xDSL in a future commit). They can also inherit MLIRType to express that they should be printed as a type rather than an attribute.

I added a unit test file, and a simple test file for LLVM. The LLVM test is quite small, mostly because I did not wanted to implement the printing for each dialect, we can do this step by step.

[webmiche]

LGTM!A short test of printing and reparsing with MLIR also worked fine.

I think you did not yet handle UnitAttrs, which I feel should be part of this PR as well.

[math-fehr]

I added differential testing in mlir-conversion/ops.xdsl !

[Dinistro]

Does this now pass all the mlir related tests or is there still something missing? Unfortunately, I do currently not have an up to date python bindings build and can thus not test it locally.

[webmiche]

It works on my machine, so LGTM!

[Dinistro]

We might consider to add an LLVM build to the CI and pin the LLVM version with a submodule. Running these test in the CI might be beneficial.