So if you have two different CSV structures, say structure-1 with columns "a","b" and structure-2 with "c","d"
then...
Can you put all files with structure-1 into a sub-folder called structure-1 and so on?
If so then you can use regex if conditionals to separate the files so that they flow through their own csv filter. You still have to take into account the second pitfall of my previous post unless you hard code the columns for each structure.
Logstash do not have an automatic way of doing this, because by design we strive for statelessness and the file input (and Filebeat) is designed to resume reading from where reached the previous time.