Add torchbench exports and benchmarks.

[monorimet]

A README.md is included in the new directory for instructions on exporting the models. Not all are functional and the process requires a clone of the pytorch source code, currently, as well as a Linux distro for pytorch+rocm support (the torchbench modeling code relies on CUDA apis)

Once the exports are covered we can set up a simple benchmarking suite that runs the compiled .vmfbs and records benchmark results.

[monorimet]

Notes on current state:

• Need to separate out runtime functionality from export script
• Start targeting torchbench integration with improvements to implementation, classing, statistics reporting, etc..
• Integrate with SHARK-TestSuite CI
• Work out upload/fetch of MLIR and weights so that this process can be used with containers running pytorch 2.3.0 without workarounds