User modifies test code while it is running, which jams the inline snapshot

1. there is testcode
2. then it gets compiled
3. then the test starts running
4. when the test completes, Selfie modifies the testcode

When selfie modifies test code, it doesn't blindly trust line numbers. It parses the literal which is present, and makes sure that it matches the value it expected before replacing it with a new value.

But just the same, all it can do right now is throw an error, and it takes all the blame. If the user modifies files between **2** and **3**, there's not much we can do about it. But if the modification happens between **3** and **4**, we should be able to say "hey don't do that!".