Add linear kernel method in kme

[Sean Kent]

It seems possible that different kernels could be useful in kme() and the code is set up decently to allow for that. The big problem is that this would require a new parameter for many of the different methods, though maybe this could get pushed into the ... argument. Sigma get's passed to it anyways

Need to think about this more. I think that all of the functions that use kme have a kernel matrix that can be passed, so in theory, this issue isn't urgent. But it would be better to tackle before the first release if it's going to get tackled.