AliasTable method is not working for unbalanced values

[s271]

Here is comparison of AliasTable and sample. sample working properly, AliasTable produce wrong result for values ~1e5 and more

using StatsBase
using Distributions
Random.seed!(1)

priorities = collect(1:1000) .* 1.
priorities[1:50] .= 1e8

isample = sample(1:1000, Weights(priorities), 16; replace=false)
_at = Distributions.AliasTable(priorities)
iat = rand(_at, 16)

println("sample ", priorities[isample])
println("AliasTable ", priorities[iat]) 

Results:
sample [1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8, 1.0e8]
AliasTable [94.0, 923.0, 388.0, 529.0, 334.0, 930.0, 630.0, 407.0, 396.0, 959.0, 215.0, 688.0, 704.0, 825.0, 524.0, 565.0]
AliasTable ignoring high-probability samples.

[simonbyrne]

The priorities need to be normalised first:

_at = Distributions.AliasTable(priorities / sum(priorities))

gives the correct result.

We should check this.