rpc can't server state based data until persisted

[advaita-saha]

When the blocks are in memory and is not persisted to the database, during that time rpc requests raise an exception due to

nimbus-eth1/nimbus/rpc/server_api.nim

Lines 60 to 66 in 9b66686

	proc ledgerFromTag(api: ServerAPIRef, blockTag: BlockTag): Result[LedgerRef, string] =
	let header = ?api.headerFromTag(blockTag)
	if api.chain.stateReady(header):
	ok(LedgerRef.init(api.com.db, header.stateRoot))
	else:
	# TODO: Replay state?
	err("Block state not ready")

One of the ways is to replay the state, but on every rpc request state getting replayed might be a bad idea.
This also directly affects the working of eth_getTransactionCount endpoint raising an exception and stopping the blob-spammer of kurtosis

[arnetheduck]

These requests need to be routed to forkedchainref which will keep a version of the state for each known unpersisted state in memory - for now, that involves a replay, but that should be easy to switch to a list of "known heads".