Retrain on failing samples only- increase from 30% -> 60% for 30 crossing 2x2 problems on benchmark. Introduce dataset01, 1000 difficult, possible sampels (#22)

* wip retrain without squares

* update weights based on training run

* max iterations based on crossings

* wip

* script for dataset generation

* Add dataset01 samples and update weights

- Add dataset01 split into 1000 individual sample files
- Update weights based on "failing only" dataset training run

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* format

* manually import samples into fixture

* wip

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

Author: seveibar

biome.json

@@ -1,12 +1,18 @@
 {
-  "$schema": "https://biomejs.dev/schemas/2.0.0/schema.json",
+  "$schema": "https://biomejs.dev/schemas/2.3.11/schema.json",
   "assist": { "actions": { "source": { "organizeImports": "on" } } },
   "formatter": {
     "enabled": true,
     "indentStyle": "space"
   },
   "files": {
-    "includes": ["**", "!**/cosmos-export", "!**/dist", "!**/package.json"]
+    "includes": [
+      "**",
+      "!**/cosmos-export",
+      "!**/dist",
+      "!**/package.json",
+      "!**/datasets/**"
+    ]
   },
   "javascript": {
     "formatter": {