ethclient: fix tx sender cache miss detection

[PlasmaPower]

types.Sender caches signers' recovered addresses for transactions. senderFromServer takes advantage of that with two different invocations:

• It can be used with both blockhash and addr specified to populate the cache. senderFromServer will return the already known address, which types.Sender will remember and attach to the transaction.
• It can be used with only blockhash specified to query the cache. senderFromServer's Equal implementation only checks the blockhash, so if the address was previously cached for this block hash, types.Sender will remember that and return it. However, if the cache misses, senderFromServer should return an error, as it doesn't know the address. That's what this PR fixes. Previously, the check for a cache miss was if blockhash was empty, which is never the case. This PR fixes that and changes the check to if addr is empty, which is the case when senderFromServer is being used to query the cache instead of populate it. Now, when that happens but types.Sender's cache misses, senderFromServer will correctly return an error.

This mechanism is used by *ethclient.Client's TransactionSender method. This bug manifested as that method returning an empty address when the cache missed, instead of querying the server as intended.

[holiman]

LGTM, I think.

[ligi]

@fjl will write a test for it