Sync committees dirtying `state.balances`

Benching a recent mainnet block

These benchmarks involve getting a recent mainnet BeaconState (slot 1783731) and then deriving two BeaconState objects:

A "clean" state with no changes.
A "dirty" state where 512 state.balances entries have been incremented by 1. The changes are evenly distributed across the state.balances array.

Then, for each state, we perform a cached tree-hash (with an already initialized cache). Results:

Clean: 35.642 ms
Dirty: 37.109 ms

Here, the difference is 1.47 ms. However, the differences vary across runs, it is generally within 1-2ms.

Caveat: cached tree hashing uses parallelization and the timings are highly dependent on the speed and availability of the CPU. So, whilst the increase here is 4%, mileage will vary. This test was conducted on a hex-core i7 running many other applications.

Source code for bench: https://github.com/paulhauner/lighthouse/blob/paul-altair-benches/consensus/types/benches/altair_balance_changes.rs