Harden fallback gateway runtime lifecycle

[8Dionysus]

Summary

• Add an explicit fallback-gateway compose profile for Ollama/LiteLLM fallback routing.
• Document compose module/profile roles, keep the llama.cpp agent sidecar outside normal profiles, and add a decision record for the fallback posture.
• Move active runtime examples away from legacy --profile core usage and teach the validator to keep them clean.
• Make aoa-warmup profile-aware: llama.cpp warmup stays local-worker oriented, Ollama fallback warmup is opt-in, and timeout handling no longer masks the fallback route.

Validation

• python scripts/release_check.py
• shellcheck -x scripts/aoa-warmup mechanics/runtime-lifecycle/parts/start-stop/aoa_warmup.sh scripts/aoa-lib.sh
• python scripts/validate_stack.py
• git diff --check
• AOA_CONFIGS_ROOT=$PWD AOA_MACHINE_FIT_AUTO_APPLY=false scripts/aoa-warmup --profile fallback-gateway

Notes

• GitHub mirror stays source-only; runtime state is still generated/applied by the documented install and sync routes.
• The fallback gateway is explicit and opt-in; the source default remains substrate.