[Feat] Add dependence structures

[lrnv]

This PR allows to relax the independecne assumptions, witht eh methodolgy of my last paper. The interface goes as follows for the test (and similarly for the estimators) :

julia> fit(GraffeoTest, @formula(Surv(time,status)~stage), colrec, slopop)
Grafféo's log-rank-type-test (Method: NetSurvival.PoharPermeMethod())
1×3 DataFrame
 Row │ test_statistic  degrees_of_freedom  p_value
     │ Float64         Int64               Float64
─────┼──────────────────────────────────────────────────
   1 │        658.779                   3  1.81926e-142

julia> fit(GraffeoTest(ClaytonCopula(2,0.3)), @formula(Surv(time,status)~stage), colrec, slopop)
Grafféo's log-rank-type-test (Method: NetSurvival.GenPoharPermeMethod{ClaytonCopula{2, Float64}}(ClaytonCopula{2, Float64}(
G: Copulas.ClaytonGenerator{Float64}(0.3)
)
))
1×3 DataFrame
 Row │ test_statistic  degrees_of_freedom  p_value
     │ Float64         Int64               Float64
─────┼──────────────────────────────────────────────────
   1 │        665.707                   3  5.72664e-144

julia> fit(GraffeoTest(FrankCopula(2,-10)), @formula(Surv(time,status)~stage), colrec, slopop)
Grafféo's log-rank-type-test (Method: NetSurvival.GenPoharPermeMethod{FrankCopula{2, Int64}}(FrankCopula{2, Int64}(
G: Copulas.FrankGenerator{Int64}(-10)
)
))
1×3 DataFrame
 Row │ test_statistic  degrees_of_freedom  p_value
     │ Float64         Int64               Float64
─────┼────────────────────────────────────────────────
   1 │        12.4529                   3  0.00598223

julia> fit(GraffeoTest(FrankCopula(2,-20)), @formula(Surv(time,status)~stage), colrec, slopop)
Grafféo's log-rank-type-test (Method: NetSurvival.GenPoharPermeMethod{FrankCopula{2, Int64}}(FrankCopula{2, Int64}(
G: Copulas.FrankGenerator{Int64}(-20)
)
))
1×3 DataFrame
 Row │ test_statistic  degrees_of_freedom  p_value
     │ Float64         Int64               Float64
─────┼──────────────────────────────────────────────
   1 │        4.77414                   3  0.189103

julia>

I also rewrote the test to use less space since we were storing things that were not necessary. The test allows allows to use EdererI or EdererII now, which is... funny, but probably useless as the theory does not hold for these.