Handoff LLVM and PyTorch updates

[ashay]

Due to a change in my job responsibilities, I am unable to continue to contribute to Torch-MLIR, so this issue records tasks that I had been previously working on and their new ownership.

1. Green commits: I will continue to post LLVM green commits every week, update MHLO to said LLVM commit, and post the MHLO green commit to the Issue LLVM green commit tracking for coordination experiment #1178.

2. LLVM updates: We need a new owner for updating the LLVM tag in Torch-MLIR based on the green commit posted in the issue referenced above.

3. PyTorch updates: These are automated through the RollPyTorch Github Action, with the exception that the Github Action itself is not scheduled. However, we can improve the RollPyTorch action's robustness by: (a) setting up a Github bot that creates a daily PR for updating the PyTorch version, shape library, and Torch ODS and (b) setting up another Github bot that approves and merges the PR if builds succeed. Creating such PRs is better than our current approach of directly committing the updates to the main branch since the PR would (a) run the full gamut of tests for PRs, (b) warm up all caches, and (c) allow unrelated concurrent merges from other PRs while the RollPyTorch action is in progress. We get around the last part by running the RollPyTorch action early in the day when there is a low likelihood of other PRs being merged.

4. [Non-Recurring Task] Build cache improvements: The PyTorch source builds currently take a long time to run because PyTorch uses pre-compiled headers, which aren't directly compatible with ccache. Issue excessive cache invalidation in ccache #1323 tracks some findings and possible next steps to make the build run faster.

cc: @silvasean

[silvasean]

@powderluv @ZihengJiang @tanyokwok @ramiro050 -- do you folks have any bandwidth on your teams for helping with the LLVM updates? I think we can make a rotation of various engineers that do the update after Ashay tags the green commit hashes.

[powderluv]

We are happy take over the PyTorch updates if anyone else wants to take over the LLVM update. We are happy to help rotate on the LLVM fixes too.
Overall having a good cadence on LLVM and PyTorch updates is crucial to the continued success of the project so we will help in anyway if there are gaps.

[tanyokwok]

@ashay Thanks for the contributions to Torch-MLIR you have made. We are glad to help with the rotation of LLVM and PyTorch updates! cc @silvasean

[ashay]

Closing because we seem to have good coverage for all mentioned tasks. Thanks folks!