How Wild Is Too Wild: Lessons Learned and Recommendations for Ecological Validity in Physiological Computing Research