Unless the csv data is in some distinct easy distinguishable format, there's no way to filter out events based on other events. Events go through ingestion pipeline one at a time and there is no global state you can rely on (to remember whether you already met the header or footer of the data part). With small chunks of this csv data you could try to do some magic involving event breaking so that input is broken on the csv header but it'd be highly ineffective and prone to errors. And you'd get a single event consisting of whole csv content, not separate csv lines and that's probably not what you want.
I'd go for scripted/modular input but that requires some development.
