docs: rebuilt README demo with measured layout; GNOME-style app icon

- demo renamed to docs/gknowledge-demo.svg (busts cached copies), AI-papers theme
- text widths measured with DejaVu Sans: no cropped labels, whole sentences only
- Adwaita-like symbolic icons, tabs centred on the content pane, proper window buttons
- new app icon: open notebook with a knowledge graph, GNOME palette and 3/4 view

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Author: francescobianco

Makefile

@@ -52,7 +52,7 @@ SANDBOX     := sandbox
 DEMO        := $(SANDBOX)/demo
 
 demo-svg:
-	$(PYTHON) build-aux/dev/make_demo_svg.py docs/demo.svg
+	$(PYTHON) build-aux/dev/make_demo_svg.py docs/gknowledge-demo.svg
 
 start:
 	@mkdir -p $(SANDBOX)/config

README.md

@@ -34,7 +34,7 @@
 </p>
 
 <p align="center">
-  <img src="docs/demo.svg" alt="gknowledge in action: drop in AI research papers, get a linked wiki, ask a question and get an answer with citations" width="880">
+  <img src="docs/gknowledge-demo.svg" alt="gknowledge in action: drop in AI research papers, get a linked wiki, ask a question and get an answer with citations" width="880">
 </p>
 
 <p align="center"><sub>

TODO.md

@@ -146,7 +146,7 @@ Tasks:
 ## Known gaps in the current MVP
 
 - [ ] Offline entity extraction is noisy on prose (for example SAN move fragments such as `Ba4 Nf6`, sentence-initial verbs, generic nouns, `Entity referenced by ingested sources.` boilerplate in answers). Filter chess/move patterns and stopword-led capitalised words, and improve extractive answers.
-- [ ] Demo animation (`docs/demo.svg`) currently shows the offline engine; re-record with an LLM provider once the ChatGPT flow is verified.
+- [ ] Demo animation (`docs/gknowledge-demo.svg`) currently shows the offline engine; re-record with an LLM provider once the ChatGPT flow is verified.
 
 - [ ] No UI for editing or deleting knowledge pages.
 - [ ] Offline heuristics are keyword-based; add language detection beyond English.