`LogitNormal` does not support zero variance

[devmotion]

Currently, LogitNormal does not allow for zero variance:

julia> LogitNormal(1.0, 0.0)
ERROR: DomainError with 0.0:
LogitNormal: the condition σ > zero(σ) is not satisfied.
Stacktrace:
 [1] #390
   @ ~/.julia/packages/Distributions/YQSrn/src/univariate/continuous/logitnormal.jl:62 [inlined]
 [2] check_args
   @ ~/.julia/packages/Distributions/YQSrn/src/utils.jl:89 [inlined]
 [3] #LogitNormal#389
   @ ~/.julia/packages/Distributions/YQSrn/src/univariate/continuous/logitnormal.jl:62 [inlined]
 [4] LogitNormal(μ::Float64, σ::Float64)
   @ Distributions ~/.julia/packages/Distributions/YQSrn/src/univariate/continuous/logitnormal.jl:61
 [5] top-level scope
   @ REPL[17]:1

whereas Normal and LogNormal do:

julia> Normal(1.0, 0.0)
Normal{Float64}(μ=1.0, σ=0.0)

julia> LogNormal(1.0, 0.0)
LogNormal{Float64}(μ=1.0, σ=0.0)

For transformed Gaussians, it should be easy to avoid issues such as #1880 since Normal already officially supports the zero variance case.