Dashboard stats could summarise scenarios rather than features

[iamam34]

Each scenario is technically a single test, so the build summary statistics should summarise all scenarios rather than all features.
Many feature files contain a mix of manual and automated scenarios, which means the reported proportion of "manual" tests is much higher than reality. Similarly, one failing scenario causes the proportion of "failing" tests to appear much higher than reality.

Discussion in HipChat room:
Current behaviour (feature-based) is useful for regression testing; this could be retained as an option.
Reporting / Printing still makes sense on a per feature basis.