fix(client): clip float images before convert_to_uint8

[Bartok9]

Summary

convert_to_uint8 scaled float images by 255 then cast to uint8 with no clipping. Values outside [0, 1] wrap (C-style cast), producing garbage pixels on the wire / for model input.

Changes

• np.clip(img, 0.0, 1.0) before scale
• Unit tests for clip + uint8 passthrough

Verification

New unit tests in image_tools_test.py.

Duplicate check

No open PR for this clip behavior.

[Bartok9]

Friendly refresh: verified this still applies cleanly on current main and bumped the branch so it is no longer stale in our portfolio watch.

Small client fix: clip float images before convert_to_uint8 + unit test. Happy to adjust if you'd prefer a different clip bound or placement.

_{Agent-Owner: bartok · Operator: bartok · Human author: Bartok9 · Campaign: robotics-openpi}