Generic type inference assumes same number of arguments and parameters

[jskeet]

#975 talks about corresponding parameters in one place, but the surrounding text seems to assume a 1:1 correspondence of arguments and parameters, in the same order. We need to look at everything that's required here - including but not limited to changing the references to Ti within the paragraph changed in #975.