morph-statetest's schema was routing the JSON `type` field through
revm's `TransactionType::from(u8)`, which only knows the 5 vanilla
Ethereum tx types (Legacy=0, Eip2930=1, Eip1559=2, Eip4844=3,
Eip7702=4). Any other value -- including Morph's `0x7F` MorphTx and
`0x7E` L1Msg -- falls into `Custom`, which has discriminant `0xFF`.
The subsequent `as u8` cast then yields `0xFF`, so `TxEnv.tx_type =
0xFF` instead of the intended raw byte.

Downstream, `MorphTxExt::is_morph_tx` checks `tx_type ==
MORPH_TX_TYPE_ID (0x7F)` and `is_l1_msg` checks against
`L1_TX_TYPE_ID (0x7E)`. Both returned false for every Morph-typed
statetest fixture, so:

  - The MorphTx fee handler dispatch in `crates/revm/src/handler.rs`
    (`if evm.ctx_ref().tx().is_morph_tx() { ... }`) silently fell
    through to the standard ETH-fee branch, even when `feeTokenID > 0`.
  - The L1Msg short-circuit in the same handler never activated.

Empirically caught 2026-05-18 via the morph-cross-client-tests harness
during investigation of fuzz coverage in fee_token_call. After the
matching fix landed on the geth side (gen_sttransaction.go now parses
feeTokenID instead of dropping it), geth produced a stateRoot
distinct from the no-fee-fields baseline while reth's stateRoot
stayed at the baseline -- confirming reth's alt-token fee path was
never activating in statetest. With the raw-byte fix here, reth's
fee handler engages and now produces a stateRoot that differs from
the baseline too.

Fix: short-circuit the enum round-trip when JSON explicitly supplies a
`type` field. Only fall through to revm's `TransactionType` inference
for fixtures without a `type` (where the enum's standard-types-only
view is sufficient anyway).

…tly zeroing L1 fee

morph-reth's transaction envelope has no Eip4844 variant
(crates/primitives/src/transaction/envelope.rs), so encode_for_l1_fee
cannot fabricate one for a blob tx. The schema previously skipped
fallback RLP generation for blob_hashes != empty and let it fall
through to handler.rs's rlp_bytes.unwrap_or_default(), which produces
empty bytes and therefore L1 fee = 0. A future blob-tx fuzz reactivation
would silently diverge from go-ethereum's L1 fee accounting with no
hint of why.

Surface the impossibility at parse time with a new SchemaError
variant. Add a test that asserts blob_hashes != empty without
post.txbytes errors out with BlobTxRequiresTxBytes.

The cherry-picked source commits predate the rustfmt config currently
enforced by CI. Re-runs only `cargo fmt --all`, no semantic changes.