…tness headers (erigontech#21529)

## Problem

`debug_executionWitness`'s `headers` field carried only the **parent
header** plus the blocks **individually** reached via the BLOCKHASH
opcode (a sparse set). A stateless verifier validates each ancestor
header by chaining `parentHash` links back from the trusted parent, so
it needs a **contiguous** chain from the parent down to the oldest
accessed ancestor — the intermediate headers are load-bearing even when
no opcode read them directly.

## Fix

`collectAccessedHeaders` now fills the contiguous range
`[oldest-accessed .. parent]` instead of emitting a sparse `{parent} ∪
accessed` set. When no BLOCKHASH gap exists (the common case), the
output is unchanged — only blocks that reach deeper ancestors are
affected.

## Evidence (EEST `zkevm@v0.4.0` corpus)

Fixes **6 of 7** `eip8025_optional_proofs/witness_headers/*` fixtures.
Confirmed expected = contiguous `[1..parent]`:

| fixture | witness block | parent | expected headers | before | after |
|---|---|---|---|---|---|
| blockhash_at_offset | 11 | 10 | 1–10 (10) | 2 | ✅ |
| blockhash_boundary | 257 | 256 | 1–256 (256) | — | ✅ |
| multiple_blockhash_max_wins | 9 | 8 | 1–8 | 2 | ✅ |
| max_wins_across_multiple_transactions | 6 | 5 | 1–5 | 2 | ✅ |
| blockhash_in_reverted_tx / _inner_call | — | — | contiguous | — | ✅ |

The 7th, `witness_headers_extra_unused_older_ancestor`, is a
**verifier-tolerance** test — its description is *"A contiguous extra
older ancestor should still validate."* It ships a reference witness
padded with one ancestor older than any BLOCKHASH access; a minimal
correct producer cannot match it by strict equality. Out of scope for a
producer-comparison suite (same category as the `witness_validation_*`
negative tests).

No regressions: the change is a no-op for blocks without a BLOCKHASH
ancestor gap. `TestExecutionWitness` passes; `make lint` clean.

Refs erigontech#21307. Independent of erigontech#21518 (witness codes from pre-state).

Co-authored-by: lupin012 <58134934+lupin012@users.noreply.github.com>

erigontech#21539)

A pre-state contract whose code hash was already produced by an in-block
CREATE is redundant in `codes` — the verifier replays the CREATE and has
the bytecode.

Track hashes written via `UpdateAccountCode`; `ReadAccountCode` skips a
pre-state read whose hash was already created. Writes flush at tx
finalization, so a read *before* the create is still kept.

Corpus (zkevm@v0.4.0): fixes `create_same_hash_then_read`, keeps
`keeps_prestate_code_read_even_if_later_created_with_same_hash`. codes
mm 11→10, no regressions.

Builds on erigontech#21518. Refs erigontech#20534.

…1531)

Order `codes` and `state` lex-ascending by content (was: codes by
codehash, state by trie-traversal). Confirmed against zkevm@v0.4.0
fixtures.

`state` is sorted after `verifyWitnessStateless` — its `RLPDecode`
treats node 0 as the trie root, so the witness stays root-first through
verification.

No pass-rate change (multiset compare); serialization alignment only.

Builds on erigontech#21518. Refs erigontech#20534 (causes 1 & 3).