README: add Quick links block — paper, honest results, live demo, home

The website was already linked from the MCU-status paragraph and the
license section, but only as a commercial pointer, not as a "here is
how to see the model run right now" link. Add a Quick-links block
near the top with PAPER.md, HONEST_RESULTS.md, the in-browser demo
at atomelm.com/demo.html, and the project home.

Also clarify "Get the kit" — weights are bundled, so a fresh clone can
run the 944K checkpoint immediately without training first.

Author: Atome LM Team

README.md

@@ -15,9 +15,17 @@ and a byte tokenizer at super-tiny scale
 ternary export → base-3 packing → C99 inference path, with bit-exact
 Python ↔ C parity enforced by tests.
 
-**Get the kit:** training code, C engine, benchmarks, paper — all in this
-repository, released under the [Apache 2.0 License](LICENSE). Train your
-own checkpoint with `scripts/train_demo.py` in ~30 min on a CPU.
+**Quick links:**
+- 📄 Architecture writeup: [`PAPER.md`](PAPER.md)
+- 🔬 Honest results, including the 944 K reversal: [`HONEST_RESULTS.md`](HONEST_RESULTS.md)
+- 🌐 Live in-browser demo (no install): [atomelm.com/demo.html](https://atomelm.com/demo.html)
+- 🏠 Project home: [atomelm.com](https://atomelm.com)
+
+**Get the kit:** training code, C engine, benchmarks, paper, and trained
+weights — all in this repository, released under the
+[Apache 2.0 License](LICENSE). Train your own checkpoint with
+`scripts/train_demo.py` in ~30 min on a CPU, or run the bundled 944 K
+checkpoint immediately.
 
 **MCU status:** QEMU ARM (Cortex-M3, MPS2-AN385) parity passes to FP32
 epsilon. **Silicon bring-up is not done in this repository.** For real