docs: add parallel sig verify documentation

Author: qj0r9j0vc2

docs/parallel-signature-verification.md

@@ -0,0 +1,66 @@
+# Parallel Signature Verification
+
+CipherBFT parallelizes ECDSA signature recovery during block execution using [rayon](https://docs.rs/rayon).
+
+## Overview
+
+ECDSA signature recovery (`secp256k1` curve) is CPU-intensive. By recovering signatures in parallel before sequential EVM execution, we reduce block processing latency.
+
+```
+Before: [Decode+Recover+Execute] → [Decode+Recover+Execute] → ...  (sequential)
+After:  [Decode+Recover ∥ Decode+Recover ∥ ...] → [Execute] → [Execute] → ...
+```
+
+## API
+
+### RecoveredTx
+
+```rust
+pub struct RecoveredTx {
+    pub tx_bytes: Bytes,      // Original RLP bytes
+    pub tx_env: TxEnv,        // Parsed for EVM
+    pub tx_hash: B256,        // Transaction hash
+    pub sender: Address,      // Recovered signer
+    pub to: Option<Address>,  // Recipient (None = create)
+}
+```
+
+### recover_transactions_parallel
+
+```rust
+impl CipherBftEvmConfig {
+    pub fn recover_transactions_parallel(&self, txs: &[Bytes]) -> Vec<RecoveredTx>;
+}
+```
+
+Recovers sender addresses from all transactions in parallel. Invalid transactions are filtered with debug logging.
+
+## Execution Flow
+
+1. Consensus delivers ordered transactions via `BlockInput`
+2. `recover_transactions_parallel()` decodes and recovers all signatures in parallel
+3. Transactions sorted by `(sender, nonce)` to prevent nonce errors
+4. Sequential EVM execution using pre-recovered `TxEnv`
+5. State committed after all transactions complete
+
+## Performance
+
+Parallelization benefits scale with:
+- Number of CPU cores
+- Transactions per block
+- Signature recovery cost (~100μs per tx)
+
+For blocks with 100+ transactions, expect 2-4x speedup on multi-core systems.
+
+## Configuration
+
+No configuration required. Rayon automatically uses available CPU cores.
+
+To limit parallelism (e.g., for testing):
+
+```rust
+rayon::ThreadPoolBuilder::new()
+    .num_threads(4)
+    .build_global()
+    .unwrap();
+```