perf: faster JPEG streaming (zero-copy VTK capture, 4:2:0 turbo-jpeg)

[user27182]

Two speedups for JPEG image streaming, one commit each:

1. Zero-copy VTK frames: VtkRemoteControlledArea captures RGBA and returns the RGB channels as a view over that buffer. The Pillow and PyTurboJPEG encoders detect the view and read the pixels in place, instead of Pillow copying every frame through Image.fromarray. img_cols_rows still returns a top-down RGB array, and output is byte-identical to master for all formats.
2. 4:2:0 for turbo-jpeg: PyTurboJPEG defaults to 4:2:2, Pillow to 4:2:0. Matching Pillow makes turbo-jpeg frames about 16% smaller and faster to encode.

Benchmark

Capture + encode per 1920x1080 frame (Apple Silicon, rendering excluded):

	jpeg (Pillow)	turbo-jpeg	turbo-jpeg frame
master	4.80 ms	4.46 ms	48.6 KB
this PR	3.96 ms (-18%)	3.72 ms (-17%)	40.8 KB

To reproduce (from a trame-rca checkout):

bench_rca.py

pip install -e . vtk
python bench_rca.py   # on master
git fetch https://github.com/user27182/trame-rca perf/rgbx-capture && git checkout FETCH_HEAD
python bench_rca.py   # on this PR

For real turbo-jpeg numbers, also pip install PyTurboJPEG with libjpeg-turbo >= 3.0 on the system; otherwise it falls back to Pillow (the script prints which one it used).

🤖 Generated with Claude Code