the `na` argument in fifelse should be coerced


This is valid:

```r
data.table::fifelse(c(T, F, NA), 1L, 2.0, 3.0)
# [1] 1 2 3
```

But not this:

```r
data.table::fifelse(c(T, F, NA), 1L, 2L, 3.0)
# Error in data.table::fifelse(c(T, F, NA), 1L, 2L, 3) : 
#  'yes' is of type integer but 'na' is of type double. Please make sure that both arguments have the same type.
```