I believe this is content length and content size being two different measurements. The limit of 85 kB is based on size that it takes up on a harddrive and the content-length of 44 k is a character length.
85 KB is the maximum size the row can take in a Coda doc file. It includes all non-formula columns and any overhead (e.g. formatted text is stored as a JSON tree of sequential pieces of content that are differently formatted). Also I can’t remember whether in Coda, text data is stored as UTF-8 or UTF-16 (i.e. whether numbers and English letters occupy 1 byte or always at least 2 bytes per character) — if it’s the latter then the size is taken up twice as fast than the size of text you send in your (usually UTF-8) API requests.
So could be either thing. Either you have extra columns that e.g. get auto-populated with data when you add a row, or Coda encodes data in UTF16 (which I honestly don’t remember, even though I used to examine the codabuf format a while ago)