Mako: one Vimba context per process; SIGUSR1 thread dump

Every failed Mako connect entered the shared VmbSystem context without
leaving it (thousands per day with a camera unplugged), so VmbShutdown
never ran and VmbC aborted at exit with a boost mutex lock_error.

- mako: enter VmbSystem once per process, leave it once via atexit
  after closing the open cameras; __del__ now closes only its camera
- app: faulthandler on SIGUSR1 (`kill -USR1 <pid>`) dumps all thread
  stacks, to find the thread that hangs the shutdown after long uptime

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>