rpc-client: Improve transaction confirmation logic for load-balanced endpoints

[joncinque]

Problem

The Rust RPC client assumes that the endpoint it's using isn't load-balanced, which can cause for false errors on transaction confirmations. For example, the confirmation logic in send_and_confirm_transaction is just checking that the given blockhash is valid for processed commitment:

agave/rpc-client/src/nonblocking/rpc_client.rs

Lines 612 to 631 in 8c7ae80

	match self.get_signature_status(&signature).await? {
	Some(Ok(_)) => return Ok(signature),
	Some(Err(e)) => return Err(e.into()),
	None => {
	if !self
	.is_blockhash_valid(&recent_blockhash, CommitmentConfig::processed())
	.await?
	{
	// Block hash is not found by some reason
	break 'sending;
	} else if cfg!(not(test))
	// Ignore sleep at last step.
	&& status_retry < GET_STATUS_RETRIES
	{
	// Retry twice a second
	sleep(Duration::from_millis(500)).await;
	continue;
	}
	}
	}

If that call hits an RPC node that's further behind, the confirmation may incorrectly return an error.

Proposed Solution

We already have the technology for this!

• getLatestBlockhash returns the last valid block height: https://solana.com/docs/rpc/http/getlatestblockhash
• getSignatureStatuses allows for a minContextSlot which gives an error when hitting a node that's behind: https://solana.com/docs/rpc/http/getsignaturestatuses

Make use of these and make the Rust RPC client a bit smarter, like the JS client.