[sharktank] Standardize Toy Llama IREE test to track regressions

[archana-ramalingam]

Standardize toy Llama IREE test to more effectively monitor regressions in the production model runs.

• Add production compile flags to toy llama test
• Move TestToyLlamaIree to GPU
• attn_head_dim must be at least 64 in toy llama

TODO:
Regenerate the inputs IDs in test_llama.py and toy_llama_test.py, as toy llama has changed

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
sharktank/sharktank/examples
export_paged_llm_v1.py
sharktank/sharktank/layers
paged_attention.py
sharktank/sharktank/layers/configs
llm_configs.py
sharktank/sharktank/models/llm
export.py
sharktank/sharktank/utils
llm_artifacts.py
llm_utils.py
sharktank/tests/models/llama
toy_llama_test.py					34-42
Project Total

This report was generated by python-coverage-comment-action