benchmark selection on the command line seems partially broken

[oconnor663]

I don't know Criterion well enough to guess at why this is, but the following behavior seems weird. In x25519-dalek and ed25519-dalek I can run benchmarks as tests like this:

cargo bench --all-features -- --test

But running the same command in the curve25519-dalek subdirectory doesn't just run tests, it runs the benchmark suite. Also, adding a filter argument (e.g. cargo bench --all-features xxx) seems to affect what tests run, but it doesn't affect what benchmarks run. Is the benchmark harness in that subdirectory ignoring CLI arguments for some reason?