For example, my service might deserialize in two different places: 1) when deserializing request bodies sent from clients, and 2) when deserializing response bodies when making API calls to other services.

I might want to enable this feature for #1 (so I can return complete error messages to clients), but disable it for #2 (because I expect this issue to be rare, and don't want to incur a performance penalty when handling deserialization errors if remote API servers give me something I don't expect).