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

[mattdawkins]

Summary

• Replace viame_train_detector with the unified viame executable across all
  platforms
• Update pipeline runner commands from kwiver runner to viame runner
• Add train subcommand when invoking training operations (viame train instead
  of viame_train_detector)
• Fix export pipeline command to use proper -p flag for pipeline path

Files Changed

• client/platform/desktop/backend/native/linux.ts - Update executable name and
  validation
• client/platform/desktop/backend/native/windows.ts - Update executable name
  for Windows
• client/platform/desktop/backend/native/viame.ts - Switch runner commands and
  training invocation
• server/dive_tasks/tasks.py - Update server-side pipeline and training
  commands
• docs/Command-Line-Tools.md - Update documentation reference

This change adapts the codebase to use VIAME's consolidated CLI interface
rather than the separate kwiver and viame_train_detector executables.

The VIAME main branch and release beta binaries have these calls already