First-level outputs

[effigies]

In #48, we switched from z-scores to effect sizes in the first-level outputs. These are more useful at the second level, but not great for visualizing.

In nilearn/nistats#183, we added the ability to easily get all output types ('z_score', 'stat', 'p_value', 'effect_size', 'effect_variance'), so which of these are the most useful? Do people have opinions?

cc @satra @mgxd

[tyarkoni]

See #125 for semi-related thoughts about naming schemes.

Is there any reason we shouldn't output everything by default? Other than disk space, what's the downside? I think most users are probably going to want to have the flexibility to peruse a comprehensive set of results (I certainly would).