Change default dimension order to (draw, chain, params...)

[sethaxen]

This PR changes the default dimension order from (params..., draw, chain) to (draw, chain, params...). While this is a breaking change, no user code that indexes and permutes dimensions using dimension names instead of indices will be impacted.

This changes the ordering from that adopted in #8. The benefits are:

• All those as the ordering adopted in Changing the default dimension order #8.
• Most analyses involve reductions over draws or draws and chains, which with column major arrays will generally be fastest with this ordering.
• Most plots involve slicing first by parameters and then by chains. So this layout is most natural for plotting
• A simple reshape forms a matrix with (sample, param), which is easily converted to a table of draws and which can be an input to MLJ.

Downsides of this approach:

• We use a different stacking dimension when combining draws vs combining chains, so we can't just recursively stack. But this is probably safer anyways.
• Combining draws and chains is in general a little slower. This is just an up-front cost though and is minor.
• The ordering is not simply the reverse of Python ArviZ's, so when a Python ArviZ user loads a netCDF saved here in Python ArviZ, the dimension order won't match those in Python ArviZ. This isn't a problem, since ArviZ supports all dimension orders.

[sethaxen]

cc @ParadaCarleton

[codecov-commenter]

Codecov Report

Merging #40 (ab61f6d) into main (e011ecb) will increase coverage by 0.98%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #40      +/-   ##
==========================================
+ Coverage   96.00%   96.98%   +0.98%     
==========================================
  Files           9        9              
  Lines         300      299       -1     
==========================================
+ Hits          288      290       +2     
+ Misses         12        9       -3     

Impacted Files	Coverage Δ
src/dimensions.jl	100.00% <100.00%> (+5.12%)	⬆️
src/from_namedtuple.jl	97.29% <100.00%> (+2.17%)	⬆️
src/utils.jl	100.00% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[sethaxen]

FYI @goedman after this PR is merged, you would need to use permute dimensions using permutedims or PermutedDimsArray in StanSample.