big_colstats using NAs as 0s with float type

[privefl]

test <- FBM(3, 1, type = "float")
test[] <- c(NA, 1, 2)
test[]
big_colstats(test)
var(c(1, 2))
var(c(0, 1, 2))

Should have errored before saying that missing values are not allowed..