Samples: Add benchmark visualization and CUDA/connect updates

[csu-bot-zivid]

Summary

Adds a new sample for visualizing ZividBenchmark CSV results, plus a
round of usability fixes and clearer sample descriptions.

New sample

• visualize_benchmark_results plots the CSV output from the
  ZividBenchmark sample, so you can inspect and compare system
  performance results. Requires the seaborn package.
• The README lists it under Maintenance and Prevention as
  "Benchmarking Your System".

Camera connection

• connect.py now accepts --serial, --ip and --hostname
  arguments, so you can point it at your own camera without
  editing the source.

GPU samples

• capture_and_process_image_with_cupy_on_cuda now wraps the
  Zivid GPU buffer directly with cp.asarray and verifies that the
  resulting CuPy array shares the Zivid device pointer, instead of
  building the CuPy array by hand.

Continuous visualization

• capture_vis_3d_in_loop_with_keypress_exit now exits reliably
  when the visualizer window is closed or 'q' is pressed, instead
  of sometimes reopening the window.

Hand-eye calibration GUI

• The pose pair selection GUI now uses detect_calibration_board
  for checkerboard detection when selecting calibration data.

Fixes and docs

• correct_camera_in_field now labels the correction estimate as
  a "dimension trueness error" instead of an "accuracy error".
• Clarified and tightened many sample descriptions across the
  CUDA, hand-eye calibration, multi-camera, transform, stitching,
  projector and basic samples, and in the README.
• Added short module-level descriptions to a few helper modules.
• Minor punctuation fixes in a few docstrings.

[blauer4]

LGTM