Looks good, playing around wit a bit. In principle changes to UI are fine with me.

One benefit though of the old version is that you can also request the median weight for a
specific age, something that occurs quite often. You do this by leaving out the weight argument:

pct_weight_for_age(age = 8, sex = "male", return_median = TRUE)

However, in the new implementation this doesn't seem possible (you have to supply a dummy value and it cannot be NA either), which is a bit counterintuitive:

pct_weight_for_age_v(age = 8, weight = 30, sex = "male", return_vector = FALSE)$P50

Can we make the weight argument optional for that use case?

Also, it seems the return_vector argument is implemented in a reversed way (return_vector=FALSE gives me the full vector).

One benefit though of the old version is that you can also request the median weight for a
specific age, something that occurs quite often. You do this by leaving out the weight argument.

Could we instead add median_weight_for_age() (etc.) functions, where the weight argument isn't present? I found the old behaviour a bit confusing and awkward, and this would be clearer UX.

Also, it seems the return_vector argument is implemented in a reversed way (return_vector=FALSE gives me the full vector).

return_vector = FALSE returns a data frame, and return_vector = TRUE returns a numeric vector. It's clearer when you have multiple input values:

Could we instead add median_weight_for_age() (etc.) functions, where the weight argument isn't present? I found the old behaviour a bit confusing and awkward, and this would be clearer UX.

Sure, that's fine too, even better. I think it's important to have that feature easily accessible.

return_vector = FALSE returns a data frame, and return_vector = TRUE returns a numeric vector. It's clearer when you have multiple input values:

I see. It's still a bit counterintuitive to me. Because if I only pass one value (a common use case), the output is just a single value while with return_vector=FALSE I actually do get a vector (or at least what looks like a vector in the console). return_numeric or return_df would be more intuitive to me.

Renamed the return_vector argument to return_numeric to avoid confusion, and added median_*_for_*() functions.

we can also remove the old data (in inst), no?

Yeah, I don't think we need it anymore. If we do this, do we also want to remove the read_who_table() function? (it doesn't look like we're using this internally anywhere)