An update on linalg on tensors

[nicolasvasilache]

@hanchung

Moving them out of IREE would be great. Most of them can be moved out easily, e.g., conv, dot, etc. I remember that the logic of lowering ops that have reduction loops is different. In IREE, we lower it to Linalg indexed_generic op or pooling ops, but in mlir-hlo repo we lower them to parallel loops. I’m happy to work together on this if we want to put them to another place.

Also ping @pifon2a who I believe has already started hacking some of this towards the MHLO repo?