[Inference/Web] Add a first-class inference CLI and documented API examples

[agorevski]

Finding

The repo documents training/evaluation and web/Python usage, but there is no first-class terminal command for decompiling bytecode and no documented curl/SSE API example.

Evidence

• README Quick Start stops at data generation, training, and evaluation: README.md:26-105.
• README CLI reference lists download_hf_contracts.py and train.py, but no inference/decompile CLI: README.md:204-242.
• Runbook inference guidance covers Python API and starting the web app, but not a terminal decompile command or curl flow: docs/runbook.md:375-443.
• src/model_setup.py has a main() that only sets up a training model and prints trainable parameters, not an inference CLI: src/model_setup.py:1623-1637.

Impact

A user can train/evaluate a model but cannot intuitively run inference from a shell, CI job, or scripted workflow. They must write Python code or reverse-engineer the web endpoint.

Recommended fix

Add a supported CLI such as uv run python scripts/decompile.py --model-path ... --bytecode ... (or a package entry point) plus README/runbook examples for both CLI and curl against /api/decompile.

Acceptance criteria

• CLI accepts --model-path, --bytecode and/or --bytecode-file, compiler metadata, and output format (json, solidity, tac).
• CLI validates input and exits non-zero with clear errors for invalid bytecode/model paths.
• CLI emits machine-readable JSON including analysis timings, function errors, and model config.
• Docs include a copy/paste CLI example and a curl/SSE example for the web API.

[agorevski]

Resolved in commit e4a1530. Validation: python -m compileall -q train.py download_hf_contracts.py src web scripts tests; uv run pytest -q (521 passed).