Avoid trampling on existing errors in defer funcs

[dmjones]

My fix for #24 introduced a bug, which this commit resolves. The problem was overwriting error information with the results from deferred functions, without first checking if an error had already been reported.

This could have the consequence of hiding errors entirely. Or, at least, replacing useful errors with less useful ones.

Thanks to @optnfast for spotting.

[optnfast]

That looks better. I don't see any other bad defers.