Update README hero image with before/after examples (#40)

## Summary
- Replace the single Telegram screenshot in the README with side-by-side
before/after training images
- Shows the visible effect of SDPO training on model output

## Test plan
- [ ] Verify images render correctly on GitHub

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **Documentation**
* Updated README images to display before and after training comparisons
side-by-side instead of a single demo image.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: kfallah

README.md

@@ -12,7 +12,9 @@
 Continual learning as-a-service (CLaaS) via self-distillation for OpenClaw. Personalize the model weights of your OpenClaw assistant using text feedback without model collapse.
 
 <p align="center">
-  <img src="assets/telegram.png" alt="Telegram demo" width="400">
+  <img src="before.png" alt="Before training" width="380">
+  &nbsp;&nbsp;
+  <img src="after.png" alt="After training" width="380">
 </p>
 
 ## Installation