Summary

• replace the MPI worker group's Simpler command/health TCP path with one L4-owned named shared-memory mailbox attached by local rank 0
• distribute ordered task and control envelopes with dedicated MPI communicators, including group, directed-rank, and per-rank payload semantics
• gather ranked results and errors before completing the mailbox request, with terminal timeout and complete mpirun process-group cleanup
• keep ordinary non-MPI Remote L3 workers on the existing TCP transport
• route full-group Global CommDomain setup through the mailbox and a separate domain communicator
• add protocol, lifecycle, compatibility, timeout, and compute/TLOAD smoke coverage

This is a stacked PR based on hw-native-sys/simpler#1623. It targets add-mpi-run so the review contains only the mailbox/MPI-dispatch commit and does not modify PR hw-native-sys#1623's history.

Testing

• named-mailbox Python unit tests: 10 passed
• changed-file header, English-only, EOF, whitespace, Markdownlint, Ruff, formatting, and Pyright checks
• Linux C++ and full Python unit suite on server 37
• single-host two-rank real mpirun integration on server 37
• server 37/35 A3 2x2 compute + global TLOAD smoke

The hardware and MPI integration items remain explicitly unvalidated until the server-37 agent returns logs.