Sharpen inference benchmark presentation

[devinnicholson]

What changed

• Replaced the 869-line chronological README with a concise research-facing overview.
• Led with the canonical Qwen 1.5B/L4 KV-cache pressure experiment and its matched control.
• Added the benchmark design, control methodology, system architecture, repository map, reproduction path, canonical artifacts, and explicit limitations.
• Removed stale learning-log language, the Week 1 status, the TODO roadmap, and the long sequence of training notes from the landing page.
• Updated the package description to reflect the current vLLM serving and scheduling work.

Why

The underlying experiments are technically strong, but the previous landing page buried the central result beneath procedural history and repetitive artifact links. The new presentation makes the systems contribution and evidentiary boundaries legible to research-engineering reviewers without overstating generality.

Validation

• python3 -m unittest discover -s tests (70 tests passed)
• git diff --check
• Verified all relative links in the new README resolve locally.

[Copilot]

Pull request overview

This PR refactors the repository’s landing-page presentation to be research-facing, highlighting the central KV-cache pressure result and providing a tighter map of the benchmark harness, methodology, and reproduction path.

Changes:

• Replaced the long chronological README with a concise overview centered on the canonical prefix-cache pressure experiment and matched control.
• Added benchmark controls, system design diagram, repository map, reproduction steps, canonical artifact links, and explicit scope/limitations.
• Updated the package description to reflect the current vLLM KV-cache / scheduling / serving focus.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
README.md	Rewrites the landing page to foreground the main result, controls, architecture, and reproducibility/artifact entry points.
pyproject.toml	Updates the project description to match the repository’s current vLLM KV-cache/scheduling/serving benchmarking scope.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.