What is a meaningful unit can be a line (e.g. sentence) a paragraph (e.g. a narrative account of linked sentences that show a process), a phrase (e.g. pointing to a particular discourse or schema), and so on.
Since the length of a "line" is a function of the program you are using (e.g., what margins you set), it was never meant to be a literal unit for coding. Instead, the idea of line-by-line coding means that you need to read and interpret every word and phrase. This is particularly important at the initial or open phase of coding, where you want too avoid preconceptions about any "particular discourse or schema."
I would go for line-by-line coding, but also making sure that the codes are not out of context. You dont want to 'ground' your ideas on things that your data is not saying. So code line-by-line, but also see the context where you generated the code from.