explorer: fix /blocks per-element state and adj; consensus: pbft cfg checksum

[Maxnflaxl]

Summary

Two bugs in the explorer's /blocks (plural) endpoint, reported by @dbadol as #2077 and #2078. Both live in Adapter::get_blocks, so they're fixed together.

Plus an unrelated one-line Rules fix, found while investigating why @dbadol's dappnet2 explorer wouldn't start on a post-HF6 build — see the third section.

#2077 — every element carries the first block's header

Regression from ba181a9, which replaced the old per-row get_block_impl() walk with a StateID/m_Number loop. The loop advances the row but never reloads the state for it, so every element after the first is rendered against the first block's Block::SystemState::Full:

sid.m_Number.v++;
sid.m_Row = _nodeBackend.FindActiveAtStrict(sid.m_Number);   // `s` left stale

Because the body is read from sid.m_Row (correct) while the header and context come from s (stale), the damage is wider than the report: on top of h and info, the totals table (Chainwork, Current Emission, Current Circulation, Size Compressed, Size Archive) and outputs[].Maturity are all wrong. Verified against mainnet — for block 27730 in a height=27729&n=4 series, emission reads 2,218,320 instead of 2,218,400 and the coinbase maturity 27969 instead of 27970. inputs and kernels were the only correct sections.

Fixed by reloading the state after each step.

#2078 — adj is ignored

adj simply never reached the backend: on_request_blocks parsed only height and n, and get_blocks had no adj parameter. What looked like "defaults to adj=1" was FindBlockByHeight → FindAtivePastHeight snapping to the next active height unconditionally, which also meant /blocks had no not-found path at all.

Fixed per the three points in the issue:

1. adj is parsed and passed through, and applies to the first block of the series only; the rest always follow upwards.
2. A vacant height with no adj returns {"found":false,"height":N}, matching /block.
3. Once the walk passes the tip, the remaining elements are filled with the same not-found objects, so the array always has n entries.

To keep /block and /blocks from drifting apart again, the first-block resolution is now shared: get_block's logic is factored into ResolveBlock(sid, s, h, height, adj) and both endpoints call it.

Rules — PBFT networks can't start after HF6

Separate from the above, and the reason the /blocks fixes can't be tested on dappnet2 right now: any build after 3c1e73f ("consensus: fees go into coinbase from HF6") refuses to start on dappnet2 and warp_dev3 with

EXCEPTION: Data configuration is incompatible: 28c7c7b263b6254a

That commit added << (uint32_t) 1 // fees in coinbase to the fork6 oracle, changing the HF6 hash. Both networks use SetForksFrom(0, 0), so all their forks sit at height 0 and their DBs are stamped with the HF6 hash — which FindFork() no longer finds, so NodeProcessor::Initialize throws. mainnet/dappnet/testnet/masternet are unaffected: they stamp HF5, which didn't change.

The rule itself never applies to PBFT — both the validation (assert(Rules::Consensus::Pbft != rules.m_Consensus) in block_validation.cpp) and the block assembly are gated on non-PBFT. Only the checksum line wasn't, so the PBFT nets break over a rule that can't fire on them. Gating it matches the existing conditionals in UpdateChecksum (CA.ForeignEnd, the Shielded.MaxIns/MaxOuts pair, the whole PBFT block).

Fork tables dumped per network, before and after:

Network	fork6 before	fork6 after
dappnet2	c94b179468bd193f	28c7c7b263b6254a (the hash in @dbadol's DB)
warp_dev3	1a430a4062b84b5d	e2fc30342d1b6d4c
mainnet	96df3f33ee02ad9e	96df3f33ee02ad9e — unchanged

So both PBFT nets start again on their existing DBs, with no resync and no coordinated upgrade, and mainnet — already past HF6 at 3,928,666 — is untouched. Chain data is unaffected either way, since the rule never applied to those chains.

Changes

• explorer/server.cpp — on_request_blocks parses adj and passes it on.
• explorer/adapter.h — get_blocks(Height, uint64_t, int adj).
• explorer/adapter.cpp — new ResolveBlock helper shared with get_block; get_blocks refreshes the state per element, honors adj for the first block, and emits not-found placeholders.
• core/block_crypt.cpp — the fees-in-coinbase value is fed to the fork6 checksum for non-PBFT consensus only.

Behavior

Request (dappnet2, PoS with gaps)	Before	After
blocks?height=2743&n=4	4 blocks, all with 2743's header	4 blocks, own headers
blocks?height=2742&n=4 (vacant)	series from 2743	[{"found":false,"height":2742}]
blocks?height=2742&n=4&adj=1	series from 2743	series from 2743
blocks?height=2744&n=4&adj=-1	series from 2795 ❌	series from 2743 ✅
series running past the tip	short array, empty blocks	n entries, tail is found:false

Two judgement calls worth a look, since the spec is ambiguous for PoS chains:

• Point 2 returns a one-element array rather than a bare object, to keep /blocks type-stable for clients. Easy to change if the bare object is preferred.
• The placeholder heights past the tip are consecutive (tip+1, tip+2, …), which only carries meaning on PoW. On a PoS chain with gaps they should be read as "no block here", not as a height prediction.

Closes #2077
Closes #2078