simplify concept of outcome type in the package

[simonpcouch]

Closes #13 and closes #11. This PR:

• Removes container(mode).
• Renames the type argument of adjust_*_calibration() to adjust_*calibration(method).

So, in the package,

• "type" refers to the sub-mode of the container, i.e. one of "regression", "binary", or "multiclass".
• "method" refers to the technique used to estimate the calibrator.

Notably, re: #13, this PR doesn't collapse adjust_*_calibration() into one function nor does it remove the option to define what is now type or method at specification time. I'm still intrigued at whether we could collapse adjust_*_calibration() into one function, but there's enough complexity there that it ought to live in its own PR.

[simonpcouch]

I've convinced myself that merging the two adjust_*_calibration() functions isn't a step forward for the interface at this point, so will leave #13 closed. :)