[quantization] Add generation step to StaticGemma4Runtime

[dvsav]

Related issue: #768
Related PR: #822

What

This PR implements the generation step (Step 8) for the StaticGemma4Runtime in the TICO project, completing the full 8-step static runtime verification flow for Gemma4 E2B. It adds a greedy generation method with side-by-side validation against HuggingFace's reference implementation.

Why

The static runtime needed a complete generation loop to verify end-to-end functionality beyond single-step decode verification. This enables:

• Full sequence generation validation (not just single-token decode)
• Token-level accuracy tracking against HF reference
• Early detection of error accumulation across multiple decode steps
• Visual inspection of generated text for qualitative assessment

Key Design Decisions

1. Greedy generation loop: The generate_greedy method reuses prefill() and decode_one() internally, ensuring consistency with the verified single-step implementations.

2. Token accuracy metric: Instead of exact sequence match (which is unrealistic for quantized vs. FP models), we report token-level accuracy percentage and first mismatch position.

3. Detokenized output: Both runtime and HF generated sequences are detokenized and printed for visual inspection, enabling qualitative assessment beyond numerical metrics.

4. No EOS enforcement in verification: The verification uses eos_token_id=None to generate exactly max_new_tokens, ensuring deterministic comparison length.

5. Cache reuse: The _raw_inputs cache from build_static_inputs is reused to avoid redundant processor calls (which involve expensive image preprocessing).

Changes

tico/quantization/recipes/debug/static_gemma4_runtime.py

• Added generate_greedy() method to StaticGemma4Runtime class (~44 lines)
• Added verify_step_generation() validation function (~88 lines)
• Updated run_static_gemma4_runtime() to execute Step 8 and print generated text

.gitignore

• Added CODE/, .vscode/, .workspace_rag/ to ignore patterns

Tests

• All 21 existing tests in test/quantization/recipes/integration/test_static_gemma4_runtime_helpers.py pass
• The verify_step_generation function performs end-to-end validation:
  • Compares runtime vs. HF generated token sequences
  • Reports token accuracy (%) and first mismatch position
  • Prints detokenized text from both paths for visual inspection

Example Scripts

Run the full 8-step static runtime verification:

python -m tico.quantization.recipes.debug.static_gemma4_runtime \
    --config tico/quantization/examples/configs/static_gemma4_runtime.yaml

Or via the inspector:

python -m tico.quantization.examples.inspector \
    --mode static-gemma4-runtime \
    --config tico/quantization/examples/configs/static_gemma4_runtime.yaml

Expected output includes Step 8:

[run_static_gemma4_runtime] Step 8: verify generation
[verify_step_generation] Runtime generated X tokens
[verify_step_generation] HF generated Y tokens
[verify_step_generation] Token accuracy: XX.XX% (X/Y)
[verify_step_generation] Runtime text: ...
[verify_step_generation] HF text: ...