[FEATURE] executor — VLA skill low-level optimization (inference accel + trajectory smoothing)

[enkerewpo]

Summary

Low-level runtime support in the Executor for VLA-based skills: inference acceleration and trajectory smoothing, wired into the Robonix API.

Goals

• For loaded VLA-based skills: inference acceleration and trajectory smoothing.
• Integrate with the Robonix API: VLA model loading, and trajectory smoothing applied before primitive calls.

Notes

Works on top of the executor's single-RTDL, switch-anytime execution model (see #52): one RTDL graph at a time, switchable on the fly.

Target window: 2026-06-08 → 2026-07 (CCF Aug-15 release).