No, a schema used is for validating an XML file, not for specifying how it might be transformed. This information would be part of the XSLT stylesheet. But
you should not have to specify where fields start in the CSV - that's what the "C" of CSV is for - it indicates field boundaries.
For this problem, I wouldn't use XML or XSLT. There are any number of libraries that can read CSV and present the information in an object model, from where you can generate the output any way you want. That seems easier than an XSLT approach (which is anyway not exactly straightforward for non-XML input).