feat: Map NA to SQLNULL

[krlmlr]

Is this a reasonable change? This helps with at least one package with duckplyr.

[krlmlr]

As it stands, this is a breaking change:

v0.10.1

library(duckdb)
#> Loading required package: DBI
con <- dbConnect(duckdb::duckdb())

out <- dbGetQuery(con, "SELECT ? AS a", params = list(NA))
dput(out$a)
#> NA

^{Created on 2024-04-24 with reprex v2.1.0}

v0.10.2 RC

...
dput(out$a)
#> NA

^{Created on 2024-04-24 with reprex v2.1.0}

This PR

...
dput(out$a)
#> NA_integer_

Any other failure modes you could think of, Hannes?

[krlmlr]

With the latest change:

...
dput(out$a)
#> NA

^{Created on 2024-04-24 with reprex v2.1.0}

Still wondering if SQLNULL is the way to go here. These are the semantics of dplyr, anyway: whenever we see a single NA, it is treated as an untyped missing value.

[krlmlr]

Need to see about the R-devel checks. Merging now, review still appreciated.