Panics are treated like exceptions, causing information leakage

When a resolver panics, the current implementation:

a. Recovers from the panic
b. Returns the error from the panic to the user in the GraphQL error list

See https://github.com/graphql-go/graphql/blob/master/executor.go#L38

Panics in go are meant to be game over, cannot continue, so if a critical issue is encountered where the program cannot continue, the last thing I want to do is continue AND leak details of the panic to my end users.

Can we at least make panic recovery optional, or ideally remove it altogether?