[tracking] Alt e2eshark TODO's

[zjgarvey]

Please feel free to contribute to this project by picking up one of the TODO items for alt_e2eshark

• (High Priority) config-specific x-fails/no-run lists.
• look into using pytest or another testing framework.
• generate command-line reproducers for ease of debugging individual stages
• Add a testing config that is identical to default config, but executes via command line scripts.
• come up with a better name for the directory than "alt_e2eshark".
• identify functionality present in e2eshark that is currently lacking and should be migrated over. E.g. Chi's comment about the --report flag.
• add more type hints to make it easier to track down definitions.
• add (much) more to the README about adding tests and running them with various command line options.
• look into removing torch-mlir as a python dependency entirely (perhaps optionally, so torch-mlir developers can still use the test suite for debugging op conversions).
• Consider alternative options for storing cache directory paths, log directory paths, onnx model paths, etc. OnnxModelInfo is currently a bit clunky in needing to sometimes infer the log directory from the model path, etc.
• Add a pytorch model info class and paths for testing pytorch using iree-turbine, or through torch.onnx.export.
• As part of adding pytorch testing configs, refactor some of the config backends.
• Look into multiprocessing
• Look into running individual tests as a subprocess, so that a hard crash doesn't tank the entire test run.

[zjgarvey]

Ci Setup:

• flesh out report feature (maybe ignore time report)
• add a report merge feature
• testfile intake / registry
• register all of the tests we want to run
• maybe setup json configurable tests (and also automate the registry of such tests)
• have a option to report with smaller stages (5 from e2eshark)