I think Machine Learning classifiers will lead. I dont have much idea about Lexicon based methods, but as per current scenrio Machine learning techniques are progressing more and more.

Dont have idea about your second question.

I think that sentiment analysis will always depend on lexicon based methods to one degree or the other. Where I see much potential is in machine learning overtaking some of the manual labor of some lexicon based tasks that are labor intensive. For example, lexicon sentiment creation is labor intensive and there are already unsupervised methods to create them. This is where I think machine learning will play a crucial role. Other examples would be using n-gram models using HMM's to make language models for sentiment analysis.