fix(harness): unique chain id per run; stop reusing stale genesis

[bdchatham]

Root cause

The nightly chains halt at height 1 and never produce blocks, so nothing becomes EVM-serving and the harness blocks → NightlyRunFailed.

Why: the cronjobs pass a static SEI_CHAIN_ID (bench, rel). Every run reuses the prior run's persisted genesis (keyed by chain id). The nodes boot with fresh validator/P2P keys, but genesis still names the prior run's validator set — so the live validators aren't the ones genesis expects, consensus can't reach 2/3 of the genesis set, and it halts at height 1 (RoundStepPropose, voting_power=0, validators dialing phantom peer NodeIDs — 5ad3366f… etc. — that exist nowhere in the live config; the real persistent_peers are correct).

Evidence (live): all bench nodes frozen at tendermint_consensus_height=1 for 5m+, 4 peers each, validator logs spamming r.runConn err="peer NodeID = <actual>, want <stale-genesis-id>".

Fix

runChainID(base) appends a per-run token to the base chain id, so each run provisions a fresh genesis + keys and never collides with prior state. benchmark/release used SEI_CHAIN_ID raw; chaos suffixed only by scenario (also collided across runs). The harness image is distroless/static (no shell), so this can't be done in the CronJob — it's derived in-process, matching the chaos suite's existing "base" semantics.

Also reverts the storage.state_store.enable=false override from #445 — that came from an incorrect "state-store deadlock" diagnosis (seid was never wedged; a goroutine dump showed it fully running — the chain was halted). State store returns to the image default.

Validation

• go vet -tags integration ./test/integration/ clean; gofmt clean.
• Post-merge: rebuild harness image → bump the platform cronjob → re-trigger load → the chain should advance past height 1, seid binds :8545, seiload runs and emits seiload_run_duration_seconds → NightlyRunFailed clears.

Supersedes #445 (wrong diagnosis). PLT-758 to be re-scoped/closed accordingly.

🤖 Generated with Claude Code

[cursor]

PR Summary

Medium Risk
Changes how nightly chains are named and provisioned in-cluster (genesis collision fix) and reverts test storage overrides; scope is integration tests only, but failures would block release/load/chaos signal.

Overview
Nightly integration suites were reusing a static SEI_CHAIN_ID, so new validator keys did not match persisted genesis from earlier runs and chains could stall at height 1. The harness now derives the effective id with runChainID, appending a nanosecond token in base36 to the env base so benchmark, release, and chaos each get a fresh genesis/key set; docs treat SEI_CHAIN_ID as a base id.

memiavlStorageConfig drops the explicit storage.state_store.enable=false override and only pins memiavl_only state commitment, leaving the state store at the image default. Comments were updated for nightly-gc label reaping and the corrected storage rationale.

^{Reviewed by Cursor Bugbot for commit c484971. Bugbot is set up for automated code reviews on this repo. Configure here.}