Test13 hardfork strategy & rollout plan

[moul]

This PR tracks the work to launch test13 as a hardfork of gnoland1, using the same tooling stack as gnoland-1 (gnogenesis fork generate + post-replay govDAO migrations).

Plan

Phase 1 — bootstrap from #5486

1. Fork PR test(hf-glue): hardfork end-to-end testbed (DO NOT MERGE) — integrates #5511 + #5376 #5486 (moul/hf-glue-experimental) as the starting point. It already has: InitialHeight fork support, PastChainIDs genesis-mode, --patch-realm, --migration-tx, post-replay valset-swap migration, full replay verified against gnoland1.
2. Adapt for test13 — swap chain IDs (gnoland-1 → test-13), point misc/deployments/ at a test-13/ directory parallel to gnoland-1/, update seed node / RPC defaults.
3. Local iteration — make reset init fetch up loop against a local gnoland-1 RPC until genesis builds cleanly and replays with 0 failures.

Phase 2 — pull in the backlog

4. Cherry-pick the ~100 commits Morgan identified for test13 (realm upgrades, VM fixes, SDK changes, etc.). Rebase them onto the hardfork branch in logical batches; each batch is a hf_patch_addpkg / code change / keeper change.
5. Local iteration again — replay + boot + smoke-test after each batch. Triage failures in situ (skip-list vs fix-forward).

Phase 3 — coordinated fork

6. Switch source from RPC to local blockstore — use gnogenesis fork generate --source /path/to/stopped-node-data (already supported via fetch-from-dir flow). This lets validators do a coordinated stop at height N and each generate a bit-for-bit identical genesis from their own local data, instead of depending on a live RPC.
7. Multi-node dress rehearsal — coordinate with the validator set to run the full stop→generate→boot flow on a shared testnet cluster. Cross-check sha256(genesis.json) across all participants.

Phase 4 — split for review

8. Once the integrated branch works end-to-end, split into three reviewable PRs:
   • PR A — hardfork logic (if anything in test(hf-glue): hardfork end-to-end testbed (DO NOT MERGE) — integrates #5511 + #5376 #5486 isn't merged yet: replay-time primitives, tooling, tests)
   • PR B — feature cherry-picks — ideally direct cherry-picks onto master, not a single PR (keeps blame/history clean; each change stands on its own)
   • PR C — test-13 chain configuration (misc/deployments/test-13/, migrations, params, seeds)

Splitting late (not early) because the integrated branch is what we actually test; the PRs are a reviewability artifact, not a development model.

Why this order

• Start from test(hf-glue): hardfork end-to-end testbed (DO NOT MERGE) — integrates #5511 + #5376 #5486, not master: we'd rebuild the same plumbing.
• Single branch through phase 3: fast iteration; split PRs would churn conflicts.
• RPC → local blockstore last: RPC is fine for development, local blockstore is mandatory for coordinated launch (no single point of dependency).

Risks / open questions

• 100-commit backlog may surface hardfork-incompatible changes; budget time for triage.
• govDAO migration txs need per-chain review (test13 validators ≠ gnoland1 validators).
• Need consensus on stop height ahead of phase 3.

Dependencies

• #5486 — hardfork tooling (parent)
• #5511 — hardfork logic primitives
• Morgan's cherry-pick list (link TBD)