Fix race condition in seen caches (#1937)

## Issue Addressed

Closes #1719

## Proposed Changes

Lift the internal `RwLock`s and `Mutex`es from the `Observed*` data structures to resolve the race conditions described in #1719.

Most of this work was done by @paulhauner on his `lift-locks` branch, I merely updated it for the current `master` and checked over it.

## Additional Info

I think it would be prudent to test this on a testnet or two before mainnet launch, just to be sure that the extra lock contention doesn't negatively impact performance.

beacon_node/beacon_chain/src/block_verification.rs

@@ -418,6 +418,7 @@ impl<T: BeaconChainTypes> GossipVerifiedBlock<T> {
         // Check that we have not already received a block with a valid signature for this slot.
         if chain
             .observed_block_producers
+            .read()
             .proposer_has_been_observed(&block.message)
             .map_err(|e| BlockError::BeaconChainError(e.into()))?
         {
@@ -472,6 +473,7 @@ impl<T: BeaconChainTypes> GossipVerifiedBlock<T> {
         // have a race-condition when verifying two blocks simultaneously.
         if chain
             .observed_block_producers
+            .write()
             .observe_proposer(&block.message)
             .map_err(|e| BlockError::BeaconChainError(e.into()))?
         {