Fix estimate_gas result differs widely from the used_gas

[boundless-forest]

We found an issue last week where the estimated gas result for some transactions was significantly higher than the actual used_gas in the receipt. See this darwinia-network/darwinia#1301 for details. After some investment, the root cause was in the proof size calculation in the record_external_operation and record_external_dynamic_opcode_cost:

1. https://github.com/paritytech/frontier/blob/master/frame/evm/src/runner/stack.rs#L989-L994
2. https://github.com/paritytech/frontier/blob/master/frame/evm/src/runner/stack.rs#L1083-L1090

In the case of not hitting the account metadata, the weight info recorder tried to record the create_contract_limit value, which is a very large amount 24576. This made it easy to trigger the transaction proof size limit and then out of gas. It fooled the binary research gas estimater to make the gas higher than before. Take https://crab.subscan.io/tx/0x313ba47341cdf5b6b5ce05cc2317ee6e3b285312b6a01d62fba4c39f51ab8c34 for example, the actual used_gas is 23135, but the estimated gas result over 160000~ This PR mainly fixes this issue. Beside this, the current record_external_dynamic_opcode_cost is very messy, I have rewritten it.

[boundless-forest]

I know you are doing some refactoring to the core evm and how the current frontier works. Would you mind to reviewing this when you have time. This is a bug fix and improvement. Maybe things will be different after your refactor, but before that, we need to solve this issue in the main branch. @sorpaas

[sorpaas]

Sorry. It just took me a while to review the other parts of your rewrites.

The EVM refactoring won't block anything here -- it'll still take a while (sorry I probably should allocate more time doing reviews). But indeed, at the end we'll be able to move away from the record_external_operation structure and implement a much nicer custom gasometer.