docs: blog edits

blog: restore plot images + fix unclosed code span in reward-hacking table

blog: add reward-hacking section (3 patched exploits + 3 by-construction invariants)

docs: drop per-component reward breakdown chart and bullets

blog: clean pass + future scope section

blog: only 3 confirmed trained systems from W&B

blog: show all 6 systems honestly; GRPO=3, SFT=6, no held-out claims

cleanup: remove writeup.md (superseded by blog.md)

blog: expand system tiers table, add extensibility section

Add SFT loss curve plot

Add blog.md: accurate systems table, SFT+GRPO plots

Fix writeup links to absolute HF URLs

Sync docs/writeup.md from latest training run

ui: status banner + endpoint guide for physix-infer cold-start

docs: explain cold small model context for SFT rationale

docs: tighten reward section into table

docs: expand reward section, remove held-out mention

Upload folder using huggingface_hub