I see what you’re saying. Your first sample transcription is not appropriate. IF you had to use Nemeth Code only for parts of column headings, and you could not use Nemeth Code for the whole table, then your second sample would be appropriate.
Because a table is not necessarily read left-to-right, top-to-bottom, the code switching rules are different from those in regular narrative. Look at the column headings separately from the data in the body of the table. An opening Nemeth Code indicator in the rightmost column heading does not carry through to the body of the table. Terminate the Nemeth there, then open it again before the body of the table, if the entire table (or most of it) requires Nemeth. Depending on the column headings, it could also be more appropriate to transcribe them all in Nemeth as well (with uncontracted words), or maybe devise a key (not ever my first choice).
Every table presents its own challenges. If you want to send a print sample of what you are looking at, I'd be happy to suggest a transcription method.