Properly free memory in `OscoreConf::new` before panicking on non-UTF-8 recipients

[pulsastrix]

In theory, the panic that occurs for non-UTF-8 recipient IDs in OscoreConf::new could be caught somewhere in the call stack and not lead to program termination.

In this case, the oscore_conf has not been properly cleaned up and causes a memory leak.
Since this is really only an edge case (most programs will terminate or abort on panic), I'll address this in a future issue.

Originally posted by @pulsastrix in #47 (comment)