Couple fixes for argmax/argmin (#87758)
Removes a wrong assert, makes min number of warps = 2 (1 for some reason generates invalid code, https://github.com/openai/triton/issues/802).
Hopefully fixes https://github.com/pytorch/torchdynamo/issues/1743, cc @jansel @mlazos @soumith @voznesenskym @yanboliang @penguinwu @anijain2305 @mreso
Pull Request resolved: https://github.com/pytorch/pytorch/pull/87758
Approved by: https://github.com/Chillee, https://github.com/soumith