Improve performance by reducing signing state contention

[ChaoticTempest]

trafficstars

Currently, SigningState is being held behind a RwLock, which can lead to contention when there are multiple operations going on (i.e. claim_oidc, reveal, signature_share). This can actually slow down another person's new_account request if there are multiple ones going on, where each are trying to acquire the write lock.

Our current SigningState looks like: https://github.com/near/mpc-recovery/blob/29b64781969f944a4aa273d181d85689e156dc52/mpc-recovery/src/sign_node/aggregate_signer.rs#L21-L24

We need concurrent key insertions is the issue here, since we have multiple entry points for modifying the keys in each of the HashMaps. But before making this optimization, we should wait until https://github.com/near/mpc-recovery/issues/233 is complete to compare easily.

one simple speed up we can do is to internalize the RwLock, such that each HashMap has its own RwLock:

pub struct SigningState {
    committed: RwLock<HashMap<AggrCommitment, Committed>>,
    revealed: RwLock<HashMap<Reveal, Revealed>>,
}

@volovyks @itegulov @DavidM-D lmk what you guys think