Indeed, the file is formatted quite well in the sense that it is very predictable.

How I would do it:

* read the file in line by line as one giant column (choose a delimiter that will never be present in the file, for example |$&|)

* use a conditional split to get rid of the lines you don't need (empty lines and header)

* implement a script transformation to parse the remaining lines. You can use the first three characters of a line to decide what to do.

* output all the columns combined into one single row. You can use STN to find out when a new row starts.