ENH: Produce all available statistical maps from each analysis unit

[effigies]

Fixes #124.
Closes #125.
Closes #133.

[effigies]

@tyarkoni @adelavega We should discuss what the appropriate outputs to be displayed at each level, and what are the appropriate ones to pass to the next level.

Currently, we are displaying and passing effect_maps at the first level, and then stat_maps at all levels higher. I'm guessing this was done for 2-level models, and what we may actually want are effect_maps at all but the final level, and then stat_maps at the final level.

But a thing to note: nistats doesn't permit using effects and variances to perform second-level models; you just get one map and you run an intercept. I believe @bthirion said he uses z-score maps for this reason, but it's not clear to me that this is equivalent.

[bthirion]

No, this is not equivalent. I'm actually working on adding mixed-effects models with known firs-level variance to nistats. I have a branch with it. Will send a PR tomorrow.

[adelavega]

From what I can tell, this closes #133. I would just suggest renaming the input names of SecondLevelModel effect_maps rather than stat_maps, as that's what's actually going into them now.

[effigies]

@adelavega This works for me on a 3-level model. Do you want to test before I merge, or just go for it?

[adelavega]

Looks good from a quick glance, so go ahead and merge and I'll test with the other upcoming changes.