The schema.ini needs to be in the directory in which the files it references resides. However, one schema.ini file can contain information for multiple files. That's as per the Microsoft link that's referenced in the article I provided. This is a limitation, not in TC, but generally in Microsoft's CSV engine.
An alternative which I have employed in the past: In the CSV file, force everything to be a string by wrapping the data element in quotes. So, even integers are strings in the CSV file. Then, in the automated test code, use the aqConvert methods to convert the data into the desired format. This gives you direct control over how the data in the CSV file is used without having to depend upon operating system settings or other external files. Even changing the registry goes out the window if you run the tests on a different machine that hasn't been altered. What I've done in the past is create different routines to retrieve the data from the files and store them in an object variable for use elsewhere. That way I have complete control of the data and it's formats.