Attention pattern visualization tweaks

[dkamm]

Hey @alan-cooney! I made some quality of life tweaks for the attention pattern visualization

1. Add a background color for the upper triangular part so I can immediately see the irrelevant part
2. Reformat the tooltip a bit to separate the index from the token
3. Force tooltip to be above current cell. Reasoning is that I want to see the whole row and forcing tooltip on left or right gets in the way.
   • (Forcing above still covers previous rows though so maybe the ideal solution is to let the user toggle the direction)
4. Make the index a suffix instead of a prefix. Indices are aligned and looks better imo
5. Allow rotation up to 90 deg on the x axis labels. I found forcing 90 deg rotation to be unpleasant