In R's data.table package is there a way to join by different keys when some condition is happening?

I have two tables that have a many to many relationship by 2 keys.

One of the tables contains NA values in one of the keys. These NA values typically appear when the other table has only one existing value for that key.

I would like to join by two keys when the value is not NA and by one single key when the 2nd key is NA

At this moment I have a two step approach but I'm wondering if there is a better way to do it.

Here my reproducible example:

library(data.table)
    set.seed(14)
    dt1 <-
        data.table(
            key1 = c("A", "A", "B", "B", "C"),
            key2 = c("A-opt1", "A-opt2", "B-opt1", "B-opt1", "C-opt1"),
            measure_1 = rpois(5, 2)
        )

    print(dt1)
#>    key1   key2 measure_1
#> 1:    A A-opt1         1
#> 2:    A A-opt2         2
#> 3:    B B-opt1         5
#> 4:    B B-opt1         2
#> 5:    C C-opt1         5

    dt2 <-
        data.table(
            id = c(1:5),
            key1 = c("A", "A", "A", "B", "C"),
            key2 = c("A-opt1", "A-opt2", "A-opt2", NA, NA),
            measure_2 = rnorm(5)
        )

    print(dt2)
#>    id key1   key2  measure_2
#> 1:  1    A A-opt1  0.0287647
#> 2:  2    A A-opt2 -0.1803785
#> 3:  3    A A-opt2 -0.3011443
#> 4:  4    B   <NA> -0.9790001
#> 5:  5    C   <NA>  1.0416423

    # This is my current two step approach

    result <- dt1[dt2, on = .(key1 == key1, key2 == key2), nomatch = 0L]
    result <-
        rbind(result, dt1[dt2[is.na(key2)], on = .(key1 == key1), nomatch = 0L][, .SD, .SDcols = names(result)])

    print(result)
#>    key1   key2 measure_1 id  measure_2
#> 1:    A A-opt1         1  1  0.0287647
#> 2:    A A-opt2         2  2 -0.1803785
#> 3:    A A-opt2         2  3 -0.3011443
#> 4:    B B-opt1         5  4 -0.9790001
#> 5:    B B-opt1         2  4 -0.9790001
#> 6:    C C-opt1         5  5  1.0416423

^{Created on 2019-11-13 by the reprex package (v0.3.0)}