Fix BalancerSOR empty response classification (#71)

We've gotten alerts about the BalancerSOR failing. Looking at the logs
this comes from gas estimation errors (division by 0). It turns out that
we are trying to turn "empty responses" (with 0 outcome tokens) into
valid solutions, because our `is_empty` check no longer correct.

Here is how an empty response from the Balancer SOR API looks like
(according to the logs):
> received HTTP response status=200 OK
body={"data":{"sorGetSwapPaths":{"tokenAddresses":[],"swaps":[],"swapAmountRaw":"0","returnAmountRaw":"0","tokenIn":"0x8afe4055ebc86bd2afb3940c0095c9aca511d852","tokenOut":"0xe3dbc4f88eaa632ddf9708732e2832eeaa6688ab"}}}

As we can see, comparing this to `Quote::default()` will not yield true
because `tokenIn` and `tokenOut` are not set to default.

This PR adjusts the check to look at swaps and swap amounts.

src/infra/dex/balancer/dto.rs

@@ -270,7 +270,7 @@ impl Quote {
     /// enough liquidity, no trading path, etc.). We don't consider this an
     /// error case.
     pub fn is_empty(&self) -> bool {
-        *self == Quote::default()
+        self.return_amount_raw.is_zero() && self.swap_amount_raw.is_zero() && self.swaps.is_empty()
     }
 }