[WIP] Feature/more tests

[kkondo1981]

See #65 .

[kkondo1981]

• Added vehicle insurance dataset with Gamma(link=log) AGLM
• Fixed residuals() to handle general case. Note that old implementation other than deviance residuals seems easy to integrated into the new implementation, but remain it for a while because I need some regression test before changing it.

[kkondo1981]

• Added a demo for French mortor TPL dataset with Poisson(link=log) AGLM.
• Simplified residuals() as written in the previous comment, because I checked new implementation doesn't change results with Poisson case.

[kkondo1981]

• Added a demo for MultinomialExample in glmnet with family="multinomial".
• Modified plot.AccurateGLM() and coef.AccurateGLM() so that they can be used with multinomial regression. Note that residual.AccurateGLM() is not modified and couldn't handle with multinomial cases properly.
• Added "demo/README.md" to show results of demos.

[kkondo1981]

• Added a demo for the banknotes dataset with family="binomial".
• Update all other demos and demo/README.md to compare AGLM with simple GLM (with cross-validation for alpha and lambda).
• Modified aglm() and cv.aglm() to cope with binomial and multinomial regressions. Also modified residuals.AccurateGLM() to cope with binomial regressions (actually there was a bug and I fixed it) , but not finished for multinomial regressions.

[kkondo1981]

Modified plot.AccurateGLM(), mainly to adjust visuals in the case of multinomial regression.
Now plot.AccurateGLM():

• draws components curves (or bars) of multiple classes in one plot instead of multiple plots,
• has the parameter to use legends to distinguish components curves (or bars) of multiple classes (use_legend),
• and has the parameters to adjusting colors of graphic components (col, resid_col, smooth_col, rug_col).

Furthermore, I disallowed to use resid=TRUE in multinomial cases, because plotting those may result slightly too complicated graphs.