…ad txs

Engine newPayload receives transactions as already-RLP-encoded bytes from
the CL. The current entry path:

  1. Decode the raw RLP bytes into types.Transaction values.
  2. Construct a *types.Block from those decoded transactions.
  3. Call HandleNewPayload -> InsertBlocksAndWaitWithAccessLists.
  4. Inside, blocksToRaw calls Block.RawBody(), which loops over
     b.transactions and calls rlp.EncodeToBytes(txn) for every tx.

Step 4 re-encodes bytes we already had at step 1. For a ~200-tx mainnet
block at ~5–15 µs per RLP encode that's ~1–3 ms of pure waste per block,
plus the alloc churn (one []byte per tx via rlp's bytes.Buffer pool).

Add Block.rawTransactions [][]byte as a caller-supplied cache, plus a
companion constructor NewBlockFromStorageWithRawTxs. When the cache is
populated (and length-matches the decoded txs), Block.RawBody() returns
the cached bytes directly. engineapi.newPayload wires the raw bytes
from req.Transactions through to the new constructor — both slices
reference the same underlying buffers, no copy.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

The rawTransactions cache added on this branch holds each tx's canonical
EIP-2718 binary encoding (the engine_newPayload wire form). RawBody()
returned those bytes verbatim, but rlp.EncodeToBytes — the path it
replaced — wraps a typed tx in an outer RLP string, so for typed txs the
cached output was shorter than before by that string prefix.

That under-counts the block size in RawBlock.ValidateMaxRlpSize, so a
block over the EIP-7934 RLP block-size limit validated as VALID instead
of being rejected (the EEST enginex RLP_BLOCK_LIMIT_EXCEEDED failures).
It also diverged the bytes persisted to kv.EthTx — copied verbatim into
snapshots — from every other ingestion path, which all store the wrapped
form.

Reconstruct the rlp.EncodeToBytes form from the cached bytes instead: a
cheap RLP-string re-wrap for typed txs, unchanged for legacy txs. The
encode-skip optimization is preserved (no per-tx field walk) while the
output is byte-identical to the non-cached path.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…t paths

The RawBody() encode-skip optimization previously reached only the engine API
path (engine_server.newPayload), so external CLs and the EnableEngineAPI
embedded Caplin benefited, but the default embedded Caplin did not:
ExecutionClientDirect.NewPayload and the historical block_collector both build
the block with plain NewBlockFromStorage and then re-encode every tx in
RawBody() during insertion.

Both already hold the binary tx encodings (body.Transactions, just decoded),
so pass them to NewBlockFromStorageWithBinaryTxs. Behavior-neutral: RawBody()
output is byte-identical (covered by TestBlockRawBodyFromBinaryTxsMatchesEncoded).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…y-no-reencode