As discussed in and closes #84. I also made it use the bench profile by default when --benchmarks is supplied, and add the --bench argument to the target args like cargo bench does (otherwise it runs unit tests instead).

For reference, you can (optionally) specify which benchmarks to run by adding them as target args:

cargo instruments -t time --benchmarks -- <my_bench_function_name>

Otherwise it'll run all of the benchmarks in the target.

Limitations:

• When there are multiple targets with benchmarks, I expect cargo instruments -t time --benchmarks to fail because it'll build multiple targets and it expects a single one (like if --benchmarks hadn't been supplied). AFAIK, cargo instruments can only run one target, so it has no analogous operation to cargo bench running benchmarks in multiple targets by default, so maybe this is OK. An alternative would be to run benchmarks in the main target by default, but this might be unexpected for users of cargo bench. Workaround is to specify a single target like cargo instruments -t time --benchmarks --bin <crate_name>. Might be nice to point users in this direction in the error message.
• I'm still unsure about the name of the flag being --benchmarks since everywhere else in cargo "bench" is used, but overloading the --bench option seems more confusing. I could see an argument for --benchmark as it looks like an action, instead of using two plurals in a row ("instruments", "benchmarks"). But --benchmarks makes it clear that it'll run all of them in the chosen target.
• If the user naively uses cargo instruments --bench without a value, the error message doesn't ask them to use --benchmarks instead. But I did add a note about it in the --help message for --bench.
• We're prepending --bench to the target args like cargo bench does, but I haven't checked the cargo bench source code to see if there are exceptions to this behavior (say, when the default test harness is disabled).
• Maybe cargo instruments --bench should imply --benchmarks? We'd probably want to use the bench profile by default, for example. I'm not using non-default test harnesses or separate bench targets though, so I haven't tested what the best behavior here should be.
• Cannot run benchmarks in lib or test targets.