[SAADabb]

Summary

consensus.yaml only tolerates dropped acks (crash/omission); nandatown has no Byzantine model, and the README explicitly disclaims byzantine_agents as "not modeled." This adds one.

• New grid_node role implementing Bracha reliable broadcast (propose → echo → ready → deliver; quorum n-f; amplify at f+1), which tolerates a Byzantine sender and Byzantine receivers at once, unlike the existing single-proposer consensus scenario.
• A Byzantine node is defined purely by an equivocation config block (which peers get which value, at every broadcast it makes) — identical protocol timing to an honest node, only the content it sends differs.
• Three scenarios, framed as grid frequency regulation, n=4, tolerance f=1:
  • grid_bft_byzantine_voter — one substation lies to everyone; the honest nodes still agree on the controller's real proposed action.
  • grid_bft_byzantine_controller — the proposer itself lies, splitting the substations; honest nodes still converge via ready-amplification.
  • grid_bft_double_fault — two Byzantine nodes at once, deliberately past f=1; two honest nodes verifiably commit different actions. Its agreement stage is expected to fail — that failure is the evidence the n>=3f+1 boundary was actually crossed, not a town malfunction.
• New grid_bft validator: checks agreement/validity among honest nodes only (Byzantine-ness is read from scenario config, never inferred from behavior).
• tests/test_sim.py's closed-world ALL_SCENARIOS/FAILING_SCENARIOS lists updated accordingly.

Test plan

• pytest tests/test_lab_grid_bft.py — 12 new tests, all passing
• pytest tests/test_sim.py — updated scenario-presence lists, passing
• Full suite: 1743 passed, 26 skipped; the 19 remaining failures are pre-existing on main (Windows symlink/filesystem-permission tests), confirmed identical with this branch's changes stashed out
• All three scenarios run end-to-end via run_lab, including bundle export and HMAC signature verification (verify_bundle(...) == [])