withdrawals, consolidations: return fee instead of excess requests from read operation

[fjl]

I'm proposing we change the contract to return the required fee instead of returning the excess requests count.
The contract now computes the fee first, and then goes on to check for calldata.

The reason is simple: the read operation is a part of the contract so that people can compute the exact required
fee that must be passed. The fee computation is not straightforward, using 'fake exponential' and some constants. Any wrapper contract would have to recompute the fee from the excess value in the same way as the system contract does. Returning the fee value directly makes the formula an implementation detail, and simplifies the wrapper:

function addWithdrawal(bytes memory pubkey, uint64 amount) private {
    assert(pubkey.length == 48);

    // Read current fee from the contract.
    (bool readOK, bytes memory feeData) = WithdrawalsContract.call('');
    if (!readOK) {
        revert('reading fee failed');
    }
    uint256 fee = uint256(bytes32(feeData));

    // Add the request.
    (bool writeOK,) = WithdrawalsContract.call{value: fee}(abi.encodePacked(pubkey, amount));
    if (!writeOK) {
        revert('adding request failed');
    }
}

Note this partially undoes the changes in #30 where I moved the read operation below write. After staring at it for a while, it seems better to have the read path first, especially since it is now so small.

[mkalinin]

My main question is the cost of the fee return, won’t it be an order of magnitude higher than the potential refund?

[fjl]

Note I'm just opening this PR to have a basis for discussion. We need to evaluate.

[fjl]

This has reached its final form now.