vla: FPO holds pi0.5 with FPO++'s fine-tuning in full (E42)

[tactino]

The pi0.5 page's FPO cell now shows E42 instead of E36's collapse.

With the rest of FPO++'s fine-tuning in place, ten FPO iterations left the two seeds at 40 and 27 of 50. The collapse E11, E14 and E36 recorded was the platform running FPO incompletely. Neither seed reaches E42's bar of 42, so the page says FPO holds pi0.5, not that it improves it.

• The cell. Status holds ("runs · intact"), score 27 of 50, E42 added to its experiments, the E42 commands.
• The clip. Re-recorded from the lower seed (fpopp-s8, iteration 10), on the same scene as the others. It fails the scene (520 steps). Where an experiment ran two seeds, the clip and the number are the lower seed's, and the page says so.
• The text. vla.md and vla.zh.md replace "a run is under way" with the result, including that s8 was drifting slowly and that past ten iterations is open.

The homepage and org README sentences ("has not made it better yet") still hold and are unchanged.

Merge after PlugRL/plugrl-server#102 (E42) and the figure PR, since the page links to experiments/e42-pi0-fpo-plus-plus.