Ordering is alphabetical, which is in case of images from 1 path/row combination equal to the acquisition date. Because you are selecting images from different path row combinations the ordering is not truly chronological and you should define the ordering of your image collection.
With respect to your second question, you'll have to use getInfo's probably:...
Yes, you can do object-based pattern recognition with the Orfeo Toolbox (OTB) plugin in QGIS. It's not a native plugin to QGIS, so you will have to download it (here) and then follow the installation instructions. Besides segmentation, there are a lot of other powerful tools within OTB that you can use, such as their machine learning algorithms for image ...
Sure, there are many projects on going using Sentinel-2 as input for land cover maps production
A few examples:
Inglada, J.; Vincent, A.; Arias, M.; Tardy, B.; Morin, D.; Rodes, I. Operational High Resolution Land Cover Map Production at the Country Scale Using Satellite Image Time Series. Remote Sens. 2017, 9, 95.
Patrick Griffiths, Claas Nendel, ...
As commented, my rasterstack of 597 ndvi images is:
create a regular daily time series object by combining data and date information
(s <- bfastts(as.vector(chitral_ROI_4), dates, type = c("irregular")))
s.d.linear <- round(na.approx(s),0)
Create time series to data frame function
time.series.to.dataframe <- function(...