Cache PyTorch source builds to reduce CI time

[ashay]

This PR contains two patches that reduce the time spent in out-of-tree
builds that reference the PyTorch source code from about 90 minutes to
about 15 minutes.

• ci: cache PyTorch source builds

This patch reduces the time spent in regular CI builds by caching
PyTorch source builds. Specifically, this patch:

1. Makes CI lookup the cache entry for the PyTorch commit hash in
   pytorch-version.txt
2. If lookup was successful, CI fetches the previously-generated WHL
   file into the build_tools/python/wheelhouse directory
3. CI sets the TM_PYTORCH_INSTALL_WITHOUT_REBUILD variable to true
4. The build_libtorch.sh script then uses the downloaded WHL file
   instead of rebuilding PyTorch

• ci: warm up PyTorch source cache during daily RollPyTorch action

This patch makes the RollPyTorch action write the updated WHL file to
the cache, so that it can be later retrieved by CI that runs for each
PR. We deliberately add the caching step to the end of the action since
the RollPyTorch action never needs to read from the cache, although
executing this step earlier in the process should not cause problems
either.

[ashay]

Here is the first run, which took 2 hours to finish. The second run finished in 15 minutes.

Caches don't seem to be shared across branches, so this CI run will perform a full build of PyTorch, but subsequent CI runs should be faster.

[powderluv]

magical. thanks

[powderluv]

Only other piece we can do is tar.gz the LLVM build so OOT CI can just download that and we done in a min or two. We can check the submodule SHA for LLVM.

[ashay]

tar.gz the LLVM build so OOT CI can just download that

That’s really neat idea! I’ll try it out in a separate PR.