This is a good question and is certainly a little confusing. The first note is that this additional universe restriction is also present in the unedited source variable (US1940A_CHILDS) behind the integrated CHBORN variable. Specifically, every case that is CHBORN==0 “N/A” has either a 99 “inappropriate” or BB “N/A” value in US1940A_CHILDS. You can also verify this by looking at the case counts on the codes tabs of CHBORN and US1940A_CHILDS.
The issue really boils down to the marital status variable MARST. As the comparability tab for MARST notes: “In the IPUMS, all persons under age 12 are coded as “never married/single” for the years 1880-1930, though these censuses did not impose a minimum age.” This rule is followed in the construction of the 1940 full count file, but when the Census Bureau released the 1940 public use microdata files, they decided to restrict the question about marital status to only those who are 14 years or older. So, this means that the CHBORN variable is correct in the both the 1% and full count files, but because the 1% sample restricts the question about marital status to only those who are 14 years or older, only such individuals will have valid values for CHBORN or the source variable US1940A_CHILDS.