See if we can get raw benchmark data into docs

.github/workflows/ci.yml

@@ -110,7 +110,22 @@ jobs:
         with:
           fetch-depth: 0
           fetch-tags: true
-      - run: ./mill site.publishDocs
+          ref: benchmark
+
+      - name: Aggregate benchmark results
+        run: ./aggregate.sh
+
+      - name: Switch to main branch
+        run: | 
+          git fetch origin main
+          git checkout main
+
+      - name: Copy benchmark results into docs generation
+        run: mkdir site/docs/_assets/benchmarks && cp benchmark_history.json site/docs/_assets/benchmarks/benchmark_history.json    
+
+      - name: Generate static site
+        run: ./mill site.publishDocs          
+
       - name: Setup Pages
         uses: actions/configure-pages@v4
       - uses: actions/upload-artifact@v3

site/docs/_docs/gotcha.md

@@ -0,0 +1,14 @@
+Despite my best efforts, there are frictions associated with cross platform initiatives. 
+
+In javascript
+
+```javascript
+const arr = new Float64Array(2)
+arr[10]
+```
+
+will result in `undefined`, and your program will continue onwards with essentially undefined future behaviuor.
+
+This is materially different to the behaviour on the JVM and native arrays, which are going to throw `ArrayOutOfBounds` or some similar exception letting you know things have gone wrong. Whilst enabling this  exception throwiung behaviour is _possible_ in scalaJS (use `Array[Double]`) , it comes with a performance impact. 
+
+It is an explicit goal of `vecxt` to be performant, and to introduce as few abstractions as possible. We make no effort to solve this problem in JS land.