Reject blocks whose block access list exceeds the gas-derived limit:
bal_items * ITEM_COST <= available_gas + system_allowance

This reverts commit 4662fde.

Align DUPN, SWAPN, and EXCHANGE decoding with the updated EIP-8024 spec
(ethereum/EIPs#11306). decode_single now uses (x + 145) % 256 instead of
branching, and decode_pair uses XOR 0x8F instead of conditional subtraction.
EXCHANGE valid range extended from 0x00-0x4F to 0x00-0x51, adding two new
swappable pairs (14,15) and (14,16).

- Subtract remaining state_gas_reservoir from gas_used in
  finalize_execution (unused reservoir is returned, not consumed)
- EIP-7702 auth refund returns to state gas reservoir for Amsterdam+
  instead of regular refund counter
- Skip Osaka TX_MAX_GAS_LIMIT cap for Amsterdam+ (EIP-8037 reservoir
  model removes the per-tx gas limit restriction)

EIP-8037 state gas for new account creation must be consumed even
if CREATE fails due to insufficient balance, depth limit, or max
nonce. Only initcode revert refunds the state gas via the snapshot
mechanism.

…tion

When an SSTORE restores a slot to zero (0→x→0), the state gas refund
must go through the normal refund counter (subject to the 1/5 cap),
not directly reduce state_gas_used. The state_gas_used must reflect
peak consumption for block-level max(regular, state) accounting per
EIP-7778/EIP-8037.

Three related fixes for state gas accounting in Amsterdam fork:

1. increase_state_gas must not mutate on OOG: Reorder to charge gas
   first, then decrement reservoir, then increment state_gas_used.
   On OOG the early return leaves all counters unmodified (matches EELS).

2. Track orphaned state gas spill in reverted children: When child
   frames consume spilled state gas then revert, state_gas_used is
   restored but the spill remains in gas_used. Track this via
   reverted_child_state_spill counter and exclude from regular gas
   derivation in refund_sender.

3. CREATE account state gas survives initcode revert: Move
   state_gas_used/reservoir snapshots to AFTER the account state gas
   charge. Per EELS, CREATE's account charge is in the parent frame
   and persists even when initcode reverts.

Two fixes for Amsterdam EIP-8037 state gas accounting:

1. CREATE collision gas excluded from regular dimension: On CREATE2
   address collision, reserved child gas is consumed but per EELS
   escrow mechanism it should not count as regular gas. Add the
   consumed gas_limit to the exclusion counter so derived regular
   gas is not inflated.

2. Amsterdam intrinsic regular gas cap validation: Add missing EELS
   check that rejects TXs where max(intrinsic_regular, calldata_floor)
   exceeds TX_MAX_GAS_LIMIT. Regular gas is capped at TX_MAX_GAS_LIMIT
   in Amsterdam (excess becomes state gas reservoir), so intrinsic
   regular cost must fit within the cap.

execution-spec-tests reorganized fixtures into per-fork subdirectories
(e.g., fixtures/state_tests/for_amsterdam/ instead of
fixtures/state_tests/amsterdam/).

Matches EELS where charge_state_gas() is called in create()/create2()
before generic_create() computes the child gas limit. Previously the
state gas was charged inside generic_create() after the 63/64
reservation, leaving the parent with ~2,796 gas remaining; when the
reservoir is 0 (gas_limit < TX_MAX_GAS_LIMIT) the 131,488-gas spill
would OOG. Moving the charge to op_create/op_create2 ensures child gas
is 63/64 of the gas remaining after state gas is accounted for.

Fixes test_selfdestruct_to_self_same_tx (all 4 variants).

- Fix validate_block_access_list_size to match EELS: the check is simply
  `bal_items > gas_limit / GAS_BLOCK_ACCESS_LIST_ITEM` with no system
  allowance or per-tx base cost adjustment.

- Fix SSTORE to charge state gas before regular gas per EELS ordering.
  At the OOG boundary, the charge order determines which gas type is
  consumed: state gas spills into gas_remaining first, then the regular
  charge may OOG with the correct state_gas_used recorded.

- Remove unused constants BAL_TX_BASE_COST, MAX_WITHDRAWAL_REQUESTS_PER_BLOCK,
  MAX_CONSOLIDATION_REQUESTS_PER_BLOCK.

Resolve merge conflicts from BAL parallel execution PR and fix
EIP-8037 gas accounting in execute_block_parallel: track regular
and state gas separately, compute block_gas_used = max(regular, state)
to match the sequential path.

Add comprehensive BAL validation for the parallel block execution path:

- validate_bal_index_zero: verify system call (index 0) entries against
  actual prepare_block effects, detecting extraneous entries by comparing
  with pre-state from store
- validate_bal_withdrawal_index: verify withdrawal/request-extraction
  index entries against actual post-withdrawal state in the DB
- validate_tx_execution: smarter PART B check comparing absent-from-BAL
  accounts with pre-execution state (system_seed/store) to distinguish
  warm-access artifacts from genuine missing entries
- validate_ordering: add storage conflict check (slot in both
  storage_changes and storage_reads)
- validate_header_bal_indices: validate BAL indices don't exceed max
  valid index for the block
- validate_bal_index_zero check for prospective gas limit

Reduces bal@v5.2.0 test failures from 43 to 6 (remaining are read-only
extraneous entries needing BAL hash comparison).

- Allow as_conversions for AMSTERDAM_INIT_CODE_MAX_SIZE const (u64→usize
  is safe for this small value)
- Replace += with saturating_add for reverted_child_state_spill to
  satisfy arithmetic_side_effects lint

Add an explicit accessed_accounts tracker to GeneralizedDatabase that
records all addresses loaded via load_account during parallel execution.
This detects extraneous BAL "pure-access" entries (accounts with no
mutations and no storage reads that were never accessed).

Also fix two edge cases:
- EXTCODECOPY with size=0 now loads the target account so the tracker
  records the access (previously early-returned before any DB read).
- Storage_read validation for selfdestructed accounts now handles both
  Destroyed and DestroyedModified statuses.

Merge main's OpcodeHandler trait refactor with BAL/EIP-8037 state gas
tracking. Resolved conflicts in 4 levm files while preserving all
Amsterdam-specific changes (state gas reservoir, SSTORE costs,
SELFDESTRUCT state gas, init code size limit, burn log rename).

Add EIP-specific Amsterdam hive tests (~1000 tests) to the PR workflow
using bal@v5.2.0 fixtures. Also bump daily hive from bal@v5.1.0 to v5.2.0.

The function compared BAL system-call entries against local re-execution
of prepare_block, which is redundant (BAL is already authenticated by
its hash in the header, and the state root validates execution). It also
broke on multi-block fixtures where block N+1's BAL references block N's
hash but the fixture was generated with a different header layout
(e.g. Prague tests running on Amsterdam fork).

… eip-8025-zkboost