Use newer viame runners for training and inference as opposed to kwiver call

[mattdawkins]

Summary

Use `viame runner` (a thin wrapper that pre-loads VIAME's environment) instead of bare `kwiver runner` for pipeline execution and ONNX export, matching the runner used by the rest of the desktop pipeline tooling.

Changes vs main

• `server/dive_tasks/tasks.py`: pipeline runs (video and image-sequence paths) and `export_trained_pipeline` now invoke `viame runner -p ` instead of `kwiver runner `.
• `client/platform/desktop/backend/native/viame.ts`, `linux.ts`, `windows.ts`: replace `kwiver runner` with `viame runner` and update `-p` syntax in the desktop equivalents.
• `docs/Command-Line-Tools.md`: doc reference updated to match.

`viame runner` is identical to `kwiver runner` from a pipeline-execution standpoint but already sources the VIAME setup script and resolves binary paths against the VIAME install, which removes a class of "binary not found" / partially-loaded-environment failures we hit on Windows and on minimal Linux installs.