Test TFLite models in iree_tests

[ScottTodd]

Motivation: add regression tests for at least compiling without errors for cases like https://github.com/iree-org/iree/issues/17902#issuecomment-2229161098

Tasks:

• Add kaggle download support to https://github.com/nod-ai/SHARK-TestSuite/blob/main/iree_tests/download_remote_files.py (using the kagglehub API, the Kaggle CLI, or curl + unzip the .tar.gz)
  • See what options there are for local caching - use our existing implementation (with MD5 hash checks?) or see what they have
• Either import from .tflite to .mlir (TOSA) offline using a script like https://github.com/nod-ai/SHARK-TestSuite/blob/main/iree_tests/onnx/import_tests.py or teach https://github.com/nod-ai/SHARK-TestSuite/blob/main/iree_tests/conftest.py how to run an 'import' stage prior to the 'compile' and 'run' stages
• Generate inputs and expected outputs for each imported model using tflite, or just test compilation
  • run with 0 inputs? skip run via config files? add test case syntax for compile only tests?

[ScottTodd]

Can convert these tests: https://github.com/iree-org/iree/tree/3f51a55daf4639473f36c42cbd9d775f2a76144e/integrations/tensorflow/test/python/iree_tfl_tests , then delete that whole integrations/tensorflow/test/ folder.

[ScottTodd]

We could also create a separate repository for each frontend test suite. iree-org/iree-tensorflow-tests? Will start with the code here I guess then figure out if moving it makes sense.