Add final model input/output examples

[tutao0123]

Summary

• add two shared-prompt examples from the completed 106M TinyStories checkpoints to both READMEs
• show AR and 64-step MDLM continuations side by side
• preserve the generated text verbatim, mark truncation explicitly, and state that examples do not replace aggregate evaluation
• include one example exposing MDLM repetition and another showing a comparatively coherent MDLM continuation

Provenance

The examples come from prompt IDs 0 and 573 in the completed fixed 1,000-prompt generation evaluation, using training seeds 1337 and 2027. Prompts are 32 GPT-2 tokens and continuations were generated for 128 tokens in BF16.

Validation

• pytest tests/test_docs.py (3 passed)
• ruff check .
• git diff --check