Benchmark feature

[JulianGCalderon]

Adds BenchBlockRange Command to Replay crate under the Benchmark features.

The command is used to measure the performance of Cairo Native without the overhead of rpc requests. Caching native compilation is not yet possible in the current blockifier version

Supersedes #12

Overview

The bench block range command has roughly three distinct steps:

1. Fetch all block range data, this includes all transactions and block context information needed to execute them
2. Executes the whole block range once, to make sure all internal data required by blockifier is cached
3. Executes (and measures) the whole block range n times. This third step won't make any rpc call as all data has already been cached.

The benchmark command ended up being quite lengthy, so I left the high level command implementation in main.rs and put the inner details in benchmark.rs

execute_tx_with_blockifier

Internally, both execute_tx_configurable and execute_tx_configurable_with_state depended on our rpc state reader, bypassing the cached state. When executing transaction for benchmarking, I needed it to be indepent of the inner state reader: CachedState<impl StateReader>. To solve this I made a new function execute_tx_with_blockifier, that does not depend on our rpc state and is just a wrapper around blockifier execution. It receives all contextual information to execute the transaction.

execute_tx_configurable_with_state could be replaced by this new function. I didn't do it in this PR to avoid having to many changes.

OptionalStateReader

To ensure that no requests are made when benchmarking, an OptionalStateReader was used instead of our StateReader. This reader can be disabled so that it panics if it is attempted to read from when benchmarking

Benchmark Result in the Server

I benchmarked a single run of block 606173 in `starknet-benchmarks'. It contains 28 transactions.

• total run time: 858.34s