Avoid negative differences in ftest

[andreasnoack]

Now that we are preparing for 2.0 anyway then I think we should change ftest to avoid the negative differences. They trigger me. They can only meaningfully have one sign so lets define them to always be positive. Maybe we'd like to enforce that the ordering of models follows a nesting structure at the same time.

julia> ftA = lm(@formula(y ~ x), df);

julia> df = DataFrame(x = randn(10), y = randn(10));

julia> ftA = lm(@formula(y ~ x), df);

julia> ft0 = lm(@formula(y ~ 1), df);

julia> ftest(ft0, ftA)
F-test: 2 models fitted on 10 observations
───────────────────────────────────────────────────────────────
     DOF  ΔDOF     SSR     ΔSSR      R²     ΔR²      F*   p(>F)
───────────────────────────────────────────────────────────────
[1]    2        8.1647           0.0000
[2]    3     1  6.1289  -2.0358  0.2493  0.2493  2.6573  0.1417
───────────────────────────────────────────────────────────────

julia> ftest(ftA, ft0)
F-test: 2 models fitted on 10 observations
───────────────────────────────────────────────────────────────
     DOF  ΔDOF     SSR    ΔSSR      R²      ΔR²      F*   p(>F)
───────────────────────────────────────────────────────────────
[1]    3        6.1289          0.2493
[2]    2    -1  8.1647  2.0358  0.0000  -0.2493  2.6573  0.1417
───────────────────────────────────────────────────────────────