Don't warn about numeric coercion in set(j=)

[MichaelChirico]

Closes #6594

[MichaelChirico]

Should we also cover the i= case here?

i is slightly more likely to have benefits from avoiding (e.g. 1e8 rows), but I think the same argument still basically applies.

If you received row_idx from out of your control (e.g. as output from another function), you'll still need to run as.integer() to overcome this warning. A wash.

[MichaelChirico]

NB: the rchk error is unrelated to the current PR, probably an upstream change (though rchk has no recent changes, maybe the docker image was updated recently, or maybe there's a new LLVM version catching new issues):

Here's a PR run without this PR applied showing the same error; the change happened in the last 5 days since that commit to master shows clean.

https://github.com/Rdatatable/data.table/actions/runs/11616342257/job/32349061905

[ben-schwen]

I would either change both i and j or none of them. I agree that it's strange that we warn about this, but having a verbose message seems appropriate. Educating ppl if they are eager to learn 😁