exp/orderbook: Use the correct asset when calculating liquidity pool disbursements

[Shaptic]

PR Checklist

PR Structure

• This PR has reasonably narrow scope (if not, break it down into smaller PRs).
• This PR avoids mixing refactoring changes with feature changes (split into two PRs
  otherwise).
• This PR's title starts with name of package that is most changed in the PR, ex.
  services/friendbot, or all or doc if the changes are broad or impact many
  packages.

Thoroughness

• This PR adds tests for the most critical parts of the new functionality or fixes.
• I've updated any docs (developer docs, .md
  files, etc... affected by this change). Take a look in the docs folder for a given service,
  like this one.

Release planning

• I've updated the relevant CHANGELOG (here for Horizon) if
  needed with deprecations, added features, breaking changes, and DB schema changes.
• I've decided if this PR requires a new major/minor version according to
  semver, or if it's mainly a patch change. The PR is targeted at the next
  release branch if it's not a patch change.

What

Flips the order in which assets are evaluated when calculating deposit requirements (or "pool expectations"). This is the actual fix:

func (state *sellingGraphSearchState) consumePool(
	pool xdr.LiquidityPoolEntry,
	currentAsset xdr.Asset,
	currentAssetAmount xdr.Int64,
) (xdr.Int64, error) {
-	return makeTrade(pool, currentAsset, tradeTypeExpectation, currentAssetAmount)
+	// How many of the previous hop do we need to get this amount?
+	return makeTrade(pool, getOtherAsset(currentAsset, pool),
+		tradeTypeExpectation, currentAssetAmount)
}

The problem is not with liquidity pool math, but rather in the way we consume liquidity pools during the orderbook search.

The other changes are to add/update test cases accordingly; refer to the commit message in 6f77891 for a thorough breakdown and justification of them.

Why

#4014 raised a valid issue about the /paths/strict-receive endpoint returning impossible paths. A follow-up investigation (comment) puts us down the right path: somehow, somewhere, assets are being swapped.

In the issue, we attempt to find an ETH -> USDC path and get results when there shouldn't be any. If we track this through the path-finding algorithm, without this fix, it claims to find the following path:

700000000000 USDC -> 11340000 BTC -> 603359 YBX -> 20400907 ETH

The first of these is true: we need 0.1134 BTC to get 70000 USDC. This is a standard offer evaluation. The second hop, though, is nonsense. The BTC/YBX pool in question has reserves of 453280/19884933, respectively. It would be impossible to get 11340000 BTC from it, since obviously 453280 < 11340000. The liquidity pool math is tested extensively (see TestLiquidityPoolMath and TestLiquidityPoolExchanges in pools_test.go), so the issue must be elsewhere.

Turns out, this 603359 number is legitimate: this is how many BTC it would take to get 11340000 YBX. That's the wrong exchange direction! When we're working backwards from the destination (as is the case of strict-receive and FindPaths), we need to see how much of the asset at the previous hop we'd need to get a payout of the current asset amount. The bug is that we find how much of the current hop we'd need to get the previous amount...

Essentially, the mistake is using the incorrect assets corresponding to X/x and Y/y in the formula from CAP-38.

This closes #4014.

Known limitations

Should we do a minor version bump?