Can we get YAXArrays to automatically parse datetime variables

It seems that YAXArrays does a great job at parsing DateTime dimensions but not DateTime variables. Here's an example that points to open data.

load packages and data

using YAXArrays
using Zarr
path = "https://its-live-data.s3.amazonaws.com/datacubes/v2-updated-october2024/S70E070/ITS_LIVE_vel_EPSG3031_G0120_X1250000_Y350000.zarr";
store = zopen(path); 
ds = YAXArrays.open_dataset(store); 

inspect mid_date dimension

ds.mid_date
julia> mid_date [DateTime("2007-07-21T02:40:09.758"), …, DateTime("2024-12-19T02:19:33.907")]

inspect acquisition_date_img1 variable

ds.acquisition_date_img1
┌ 22242-element YAXArray{Float64, 1} ┐
├────────────────────────────────────┴─────────────────────────────────────────────────────────────────── dims ┐
  ↓ mid_date Sampled{DateTime} [DateTime("2007-07-21T02:40:09.758"), …, DateTime("2024-12-19T02:19:33.907")] Unordered Irregular Points
├──────────────────────────────────────────────────────────────────────────────────────────────────── metadata ┤
  Dict{String, Any} with 6 entries:
  "units"         => "days since 1970-01-01"
  "calendar"      => "proleptic_gregorian"
  "name"          => "acquisition_date_img1"
  "standard_name" => "image1_acquition_date"
  "description"   => "acquisition date and time of image 1"
  "_FillValue"    => NaN
├─────────────────────────────────────────────────────────────────────────────────────────────── loaded lazily ┤
  data size: 173.77 KB

Notice that acquisition_date_img1 has not been converted to DateTime

ds.acquisition_date_img1[1]
julia> 13499.111201768137

Is there a way that we could get YAXArrays to automatically parse DateTime variables?

Start at the YAXArrays.open_dataset entry point and trace how CF metadata such as units and calendar is applied to dimensions versus variables. Reproduce the provided Zarr example and add a regression test for acquisition_date_img1 using its days-since-epoch metadata. Done means eligible datetime variables are exposed as DateTime values without breaking numeric variables.