Non-zero WT behavior subtly breaks guide quality model

[tlnagy]

An email from a user, made me think of a potential edge case bug that could happen. Setting the "inactive/WT" phenotype to something other than 0 might lead to subtly incorrect behavior elsewhere. For example, the guide quality model assumes that 0 is the inactive phenotype:

julia> using Crispulator

julia> facs = FacsScreen();

julia> facs.num_genes = 1;

julia> max_phenotype_dists = Dict{Symbol, Tuple{Float64, Sampleable}}(:inactive => (1.0, Delta(-0.25)))
Dict{Symbol, Tuple{Float64, Sampleable}} with 1 entry:
  :inactive => (1.0, Delta(-0.25))

julia> lib = Library(max_phenotype_dists, CRISPRn());

julia> lib.knockdown_probs.p .= [0.5, 0.5] # high fraction of "low quality" guides
2-element Vector{Float64}:
 0.5
 0.5

julia> construct_library(facs, lib)[1]
25-element Vector{Crispulator.Barcode}:
 Crispulator.Barcode(1, 1.0, -0.25, NaN, :linear, :inactive, -Inf)
 Crispulator.Barcode(1, 1.0, -0.25, NaN, :linear, :inactive, -Inf)
 Crispulator.Barcode(1, 0.03168858098146547, -0.007922145245366368, NaN, :linear, :inactive, -Inf)
 Crispulator.Barcode(1, 1.0, -0.25, NaN, :linear, :inactive, -Inf)
 Crispulator.Barcode(1, 1.0, -0.25, NaN, :linear, :inactive, -Inf)

Guide 3 is a low quality guide, but theoretical phenotype is "better" because it is close to 0 instead of -0.25 like expected. The proper fix is to make "low quality" guide to behave like the inactive/WT class.