fix(devnet): green up everlight devnet test on master

[mateeullahmalik]

Summary

make devnet-tests-everlight is failing on master (7ca770a, PR #113). This PR makes it green with a single, surgical, multi-file fix.

Result on a fresh 5-validator devnet:

  PASS: 39   FAIL: 0   SKIP: 4
  EXIT=0

The 4 SKIPs are intentional (S3.5–S3.7 payout-timing on devnet, S4 longer state setup, S9 pre-Everlight upgrade flow, S10 full lifecycle).

What was broken

1. devnet/go.mod pinned lumera v1.10.0 but tests/validator/lep5_test.go (PR LEP 5 #103) imports x/action/v1/merkle which only exists at HEAD — go mod tidy failed.
2. everlight_test.sh had 9 cmd; rc=$? sites that swallow non-zero exits silently under set -euo pipefail, killing the script before any FAIL line could print (this is why the original symptom was S2.1 PASS … make: *** Error 1 with nothing in between).
3. audit_current_epoch_id used .epoch_id // empty — proto3/gogoproto omits zero-valued scalars, so a legitimate epoch 0 response renders without .epoch_id and the function returned a missing-key error.
4. audit assigned-targets --supernode-account FOO and supernode report-supernode-metrics --validator-address FOO use positional args, not flags. The script's flag form silently failed.
5. S7.7 attempted to gov-update epoch_length_blocks. The audit module correctly enforces this as immutable-after-genesis (consensus-critical epoch math). The test was wrong, not the chain.
6. The supernode's host_reporter (PR #284) auto-submits MsgSubmitEpochReport every ~5s with the SN's reporter key. The test tries to submit using the same key from outside the SN and loses the account-sequence race every time → S2.3, S2.4, S2.6, S5.x all silently failed in the recovery loop.
7. S5.4 asserted "remains eligible after growth jump". The chain correctly clamps the spike via smoothing + growth-cap, so eligibility flips false during ramp-up — the assertion races the very feature it's testing.

What this PR changes

• devnet/go.mod — activate the local replace lumera => .. block so devnet builds against this repo's HEAD; bump baseline require to v1.11.1.
• devnet/default-config/devnet-genesis.json — set epoch_length_blocks: 20 so devnet has fast (~20s) epochs. No chain-code change required; matches the design that this is a genesis-time decision.
• devnet/generators/docker-compose.go — emit EVERLIGHT_TEST_TARGET=1 only on supernova_validator_1 (the test target).
• devnet/scripts/supernode-setup.sh — when EVERLIGHT_TEST_TARGET=1, forward LUMERA_SUPERNODE_DISABLE_HOST_REPORTER=1 so v1's host_reporter is suppressed and the test can drive MsgSubmitEpochReport cleanly. The other 4 validators keep host_reporter running, so peer reachability observations still flow and ACTIVE-state eligibility still works.
• devnet/tests/everlight/everlight_test.sh — 7 fixes: set -e hardening, jq epoch-0 fix, two flag→positional fixes, S7.7 immutable-by-design refactor, S2 recovery preamble, S5.4 anti-gaming property assertion rewrite.

Companion change (required)

LumeraProtocol/supernode#285 — adds the LUMERA_SUPERNODE_DISABLE_HOST_REPORTER env knob. Without it the EVERLIGHT_TEST_TARGET=1 propagation in this PR is a no-op and S2/S5 will still fail.

No chain code touched

Every suspected "chain bug" turned out to be correct protocol behavior (epoch_length immutability, missing-report POSTPONE, host_reporter auto-submit, anti-gaming smoothing). This PR only updates devnet config, devnet test infrastructure, and the test script itself. x/audit, x/supernode, x/action, ante handlers, keepers, migrations — all untouched.

Risk / rollback

• Production risk: zero. All changes are inside devnet/ and devnet-only configuration. MinCascadeBytesForPayment, MinCpuFreePercent, epoch_length_blocks semantics, and the audit/supernode modules are unchanged.
• Rollback: revert this single commit.

Test evidence

=== Scenario 2: STORAGE_FULL State Transition (F12, F13) ===
  PASS: S2.1 resolved service supernode (validator=lumeravaloper1apws… state=SUPERNODE_STATE_POSTPONED)
  PASS: S2.3 submit audit epoch report with high disk usage
  PASS: S2.4 audit report transitions supernode to STORAGE_FULL
  PASS: S2.5 submit audit epoch report with healthy disk usage
  PASS: S2.6 audit report recovers supernode from STORAGE_FULL to ACTIVE
  PASS: S2.7 submitted legacy supernode metrics with high disk
  PASS: S2.8 legacy metrics path does not mutate state
…
  PASS: S5.4 anti-gaming smoothing clamps growth jump (smoothed=277104 < raw=21474836480)
  PASS: S5.5 smoothed_weight exposed via eligibility query
------------------------------------------------------------
  PASS: 39   FAIL: 0   SKIP: 4
============================================================
EXIT=0

[roomote-v0]

Rooviewer   See task

Reviewed all six changed files. The fixes are well-targeted and correct:

• The go.mod local replace directive keeps devnet builds in lockstep with HEAD, fixing the missing x/action/v1/merkle import.
• The set -e hardening (rc=0; cmd || rc=$? pattern) is the right idiom for capturing exit codes without tripping pipefail.
• The epoch_id // 0 fix correctly handles proto3 zero-value omission, with the epoch_start_height guard preventing silent acceptance of error payloads.
• Both flag-to-positional fixes (assigned-targets, report-supernode-metrics) match the AutoCLI definitions ([supernode-account], [validator-address]).
• S7.7 conversion from a doomed gov-update attempt to a presence check aligns with the immutability enforcement in x/audit/v1/keeper/msg_update_params.go.
• The S2 recovery preamble and S5.4 anti-gaming assertion rewrite correctly address the host_reporter race and growth-cap clamp timing issues respectively.
• The EVERLIGHT_TEST_TARGET / LUMERA_SUPERNODE_DISABLE_HOST_REPORTER plumbing through docker-compose and supernode-setup is clean and scoped to the single target validator.

No issues found.

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}