feat: add PicoVLA and update related components

[nguyentruongchinh04z]

What

• Add PicoVLA inference support with DINOv3 ConvNeXt-T, task-aware visual compression, Llama backbone, temporal memory, and MeanFlow action expert.
• Implement LeRobot-to-GGUF conversion and model loading.
• Extend C/Python/gRPC APIs with reset_memory (ABI v2) for episode-level memory management.
• Integrate PicoVLA into LIBERO evaluation with cross-chunk action blending.
• Update documentation and add converter unit tests.

Why

• Enable efficient PicoVLA inference through vla.cpp.
• Support stateful inference and reproduce the reference policy's execution behavior on LIBERO.
• Extend architecture coverage while maintaining compatibility with existing models.

Verified

• Builds clean under -Wall -Wextra (first-party code)
• ctest passes
• Numeric output unchanged (vla_predict_check diff), or the change is meant to move it and a LIBERO sweep is below

Archs and backends tested:

• PicoVLA — testing pending.