feat[4]: CLI stages, end-to-end pipeline, README, and smoke test

Goal

Iteration 4 of the SPADE implementation: wire everything together into usable command-line stages (train teacher, prune via WLI, distill, evaluate), a one-command end-to-end pipeline, configuration files, a full README, and a fast end-to-end smoke test.

Acceptance

• Stage CLIs exist: python -m spade.train, python -m spade.prune, python -m spade.distill, python -m spade.evaluate, python -m spade.pipeline
• YAML configs for the built-in benchmark + a tiny smoke configuration
• Pipeline produces a JSON report (WLI/retained indices, WER before/after distillation, params/depth/RTF comparison)
• README documents the method, repo layout, quickstart, and extension points
• End-to-end smoke test (train -> WLI -> prune -> distill -> evaluate) passes under pytest

Notes

Iteration 4 for: SPADE - Structured Pruning and Adaptive Distillation for Efficient LLM-TTS (arXiv:2509.20802)

Start by inspecting the existing Python package and the interfaces needed by the five requested module entry points. Run the current pytest suite, then verify each stage and the one-command pipeline against the benchmark and tiny YAML configurations. Done means the smoke test passes, the pipeline emits the required JSON report, and the README documents the requested usage and extension points.