Use CFtime in stars for time management of netCDF files

[pvanlaake]

This is a new PR after updating code to address issues with #690. This PR has been synched with the main branch minutes ago.

I have made a few updates to the code after studying issues with the tests. There are changes in dimensions.R and aggregate.R, compared to the changes in the previous PR.

Please note that I had to downgrade the dependency on sf: version 1.0-19 is not installing from CRAN (for MacOS) and building locally is impossible because I do not have the GDAL dev environment locally. I do not think that there is any impact of that on this PR.

Considering that there remain differences in the tests compared to the saved output, I have installed a clean version of the stars source from a zip file download from GitHub 2 days ago. Comparing tests output between that version and my PR reveals the following:

• gridtypes.R, raster.R, stars.R have diffs not related to the PR.
• area.R gives 15 warnings in both the PR code and the current main stars branch. Is area.Rout.save out of date with only 12 warnings?
• Similarly with curvilinear.R.
• dimensions.R will always have some differences because there is code added to deal with CFtime specifics. Otherwise the differences have been resolved.
• mdim.R The objects created are the same, there are only differences in the attributes which are related to the fact that a CFtime instance is used to represent time from the netCDF file when doing read_mdim().

[pvanlaake]

@edzer This PR fails on building three of the vignettes, with an error I haven't seen before. The issue is the following:

(stars1.Rmd, line 217) When making a stars object from 9 AVHRR images the result has 36 attributes named like "1.sst", "1.anom", "1.ice", "1.err", ... , "9.err". The "time" dimension is not merging and in the resulting object only the first day of 9 is included. (The error in vignettes 3 and 7 appears to be of the same nature.)

I looked around in the code, specifically in c.stars() and st_as_stars.list() and everything seems to be going fine in the first 9 iterations, but then in a 10th iteration (?) the 36 attributes pop up here:

if (identical_dimensions(dots, tolerance = tolerance)) {
    st_as_stars(setNamesIfnn(do.call(c, lapply(dots, unclass)), nms),
                dimensions = st_dimensions(dots[[1]]))
} else {
      ...

I have been trying to figure out where the code leads (nested functions, do.call(), lapply(); it's not easy to debug!) but I've come up short. Can you tell me which function(s) I should be looking at to merge the "time" dimension of the individual objects such that they merge nicely to 4 attributes? Thanks in advance for your help.