analysis: raise instead of NaN for empty particle types

[RudolfWeeber]

check_particle_type only verifies 0 <= p_type <= max_seen_particle_type,
and max_seen is never decremented when particles are removed. A
seen-then-deleted type (or any populated-then-emptied type) therefore
passes validation while contributing an empty buffer / zero total mass.

On such a type the analysis routines divided by zero:

• center_of_mass returned com/mass = 0/0 = NaN (the double mass = 1.
  placeholder was a red herring: boost::mpi::reduce overwrites the root's
  output with the reduced sum, so it never prevented the division).
• gyration_tensor divided by buf_pos.size() == 0; its non-empty
  precondition was only a release-stripped assert(), so it produced NaN
  in Release and aborted in RelWithAssert.
  moment_of_inertia_matrix inherits center_of_mass, so it threw too.

Fix (PREVENT, matching the subsystem convention where bad analysis
preconditions throw):

• center_of_mass: all_reduce the total mass to every rank and throw a
  std::runtime_error when it is zero. The throw is collective so the
  script interface does not deadlock on the following MPI collective.
• gyration_tensor: replace the assert with a real throw on rank 0.
• Analysis.cpp: move the core calls and their MPI collectives inside
  context()->parallel_try_catch (mirroring calc_rg) so the new throws
  are marshalled correctly across MPI.

Co-Authored-By: Claude Opus 4.8 noreply@anthropic.com

🤖 Generated with Claude Code