Move pull and trunk slow tests to periodic (#98040)
I notice that we are running some slow tests for CPU and `sm86` on pull and trunk. They take much longer to run than other shards (1.5x to 2x longer). I propose that we move them to periodic instead. Thoughts?
The correlation between them are:
* `linux-bionic-cuda11.7-py3.10-gcc7-sm86 / test (slow)` and `linux-bionic-cuda11.7-py3.10-gcc7-sm86 / test (default)` is 0.93
* `linux-bionic-py3.8-clang9-slow / test (slow)` and `linux-bionic-py3.8-clang9 / test (default)` is 0.98
<!--
copilot:summary
-->
### <samp>🤖 Generated by Copilot at db56750</samp>
This pull request updates the `.github/workflows` files to optimize the testing workflows for PyTorch. It adds new periodic workflows for more platforms and configurations, and removes some redundant or slow workflows from the pull and trunk workflows.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98040
Approved by: https://github.com/malfet
