HDGL UNIFIED BOOT — one image, three discardable shells
=======================================================
BOOT ORDER (all from one HDGL image, /init):
  1. phi_pool settles substrate to phi        (the core, ticks forever)
  2. Temple pane   : HDGLAgent.HC + PaneBus.HC (ExePutS live-mutation surface)
  3. Linux sidecar : emerge-sidecar.sh          (stable brain, booted by init)
  BUS: /lattice/slots (Delta) + /lattice/bus/{to,from}_bot (HolyC exchange)

THE LOOP (mutable-without-reboot, two mutabilities separated):
  world/pane -> /lattice/bus/to_bot -> sidecar asks stable LLM brain
             -> emits HolyC -> /lattice/bus/from_bot -> pane ExePutS's it LIVE
  Brain never mutates itself. Temple mutates freely. Each is strong at its job.

THE THREE SHELLS (borrowed now, discarded by the roadmap — nautilus, not molt):
  SHELL 1  LLM  : LM Studio/Ollama  -> discarded when Trailblaze glyph-model
                                        settles alone (critic glove trained)
  SHELL 2  GPU  : traditional CUDA  -> discarded when analog substrate (onion-
                                        shell HW) IS the parallel compute
  SHELL 3  OS   : Linux sidecar     -> discarded when Temple-native bot stands
  Each seam marked [SHELL n] in emerge-sidecar.sh; cut independently.

FILES:
  init.sidecar        : /init with sidecar hook (+ expanded memory from before)
  emerge-sidecar.sh   : the substrate-booted brain (CPU now, GPU seam ready)
  PaneBus.HC          : Temple side — reads bus, ExePutS live
  HDGLAgent.HC        : the self-modifying HolyC agent (earlier)

PHASE-1 STATUS (honest): CPU brain via LM Studio, substrate-booted, live
  mutation working. NOT YET: GPU/CUDA (needs driver in rootfs), Trailblaze
  glove (Phase 2), native glyph-model (Phase 3). Nothing here blocks those —
  every shell has its successor's seam already cut.
