- add initial EIP-1559 transaction validation
- add initial EIP-1559 transaction execution
- rename gas_target to gas_limit
- add london VM in existing tests
- update block according to latest EIP updates

- prefer exceptions for base gas fee to None, where feasible
- update mainnet block number for London
- increase default gas price in tests to 10 gwei to easily out-price the 1559 min gas fee of 1gwei
- make old transactions understand max_priority_fee_per_gas and
    max_fee_per_gas (just return the gas_price, as the 1559 spec suggests)
- validate the transaction nonce on new london transactions
- enable creating a genesis London header (was hard-coded to pre-London-style)
    generally allow create_header_from_parent to accept a None parent, for genesis creation
- use the correct VM's block header when creating a temp block for txn evaluation
- update tests that were configured with a gas price of < 1 gwei
- Drop support for Header.from_parent which is essentially always wrong.
    Should use vm_class.create_header_from_parent instead.
- Make eth_call free again
  - when running a call, set the header's base gas fee to 0
  - API update: state_in_temp_block -> in_costless_state to reference ^
- Correctly load uncles across fork boundary
  - Add test to validate uncles across chain VMs
  - Use a universal header sedes, since we can't load the VM before decoding the header
  - Decode block bodies with both London and pre-London uncles
  - Bonus bugfix: must double the header's gas limit at the VM boundary
- Implement base fee adjustment, with tests
  - target must be half of limit
  - gas_used_delta bug when parent_gas_used > parent_gas_target: was
    subtracting base fee instead of target
- Apply gas limit rules correctly
    Note that the 'diff' approach had a bug when 'diff' was negative. That
    approach needed a math.abs() to correctly validate the limit.

The alternative is to modify create_header_from_parent to accept and
assign all the rest of the header fields. Since none of these values are
important to customize, I'm going with this easier/smaller change.

Direct invocation of the header object is not allowed anymore, because
different VMs now have different header types (as of EIP-1559 in
London). So we must always invoke header creation through the VM.

- Remove the BlockHeader() init in test_blockchain.
- Remove HeaderChain, which seems to be completely unused in the code
    base.
- Embrace that we need one global type that can decode all headers.
- Add VM.create_genesis_header
    The genesis header now needs to be VM-aware, to handle EIP-1559. Rather
    than pass in a parent of None, it looks a bit cleaner to use the new
    create_genesis_header.
- Create header from VM in opcode test (we need a proper London header
    at genesis)

Actually look at the clock, in UTC, and compare it to the parent. Related:

- Fix doctests: set timestamp manually for reproducible results (new
MiningChain API)
- clique should verify timestamp in UTC (bug uncovered)
- Only guess timestamp on non-genesis headers (when parent & VM are
unavailable), so manually set in a test where headers are built manually

- Test London validation of sender sufficient funds

Each transaction object should now how to display itself, whether it's
a legacy, access list, or dynamic fee transaction.

- add tests to verify
- Push backwards-compatible usages of get_tip() to Frontier
- Bonus fix: setting coinbase when mining
  The coinbase address was not correctly receiving transaction fees when
  the address is passed as a kwarg to MiningChain.mine_all or .mine_block
- Bonus fix: Correctly decode dynamic-fee txns in London

This seems to be a commonly used name, and is definitely better than
BaseGasFee.

The old test implementation was finding other problems, like a
base_fee_per_gas that did not match the parent.

It just didn't show up because this is the first time that the VM header
validation rules vary so significantly.

Not really getting anything useful from crashing there. It's an ugly
crash that should be caught somewhere else.

Otherwise, get an infinite loop, where test hangs.