Pyramid from Raster seems to be faster than from YAXArray

When I am computing the Pyramid for a Raster this seems to be roughly a factor of two faster than computing the pyramid from a YAXArray with the same underlying data. This is with the example data "crs_enabled_zarr_and_tif/some.tif" from the geozarr interoperability issue. I am not sure where these differences come from and whether this is something that is worth looking into.

These are the timings I get for computing the pyramid for the first band.

julia> @time pyrc = PS.Pyramid(b1c)
  2.899440 seconds (44.07 M allocations: 1.908 GiB, 15.17% gc time)
╭──────────────────────────────╮
│ 8192×8192 Pyramid{Float32,2} │
├──────────────────────────────┴─────────────────────────────────────── dims ┐
  ↓ X Sampled{Float64} 400000.0:10.0:481910.0 ForwardOrdered Regular Points,
  → Y Sampled{Float64} 6.0e6:-10.0:5.91809e6 ReverseOrdered Regular Points
├────────────────────────────────────────────────────────────────────────────┴───────────────── metadata ┐
  Dict{String, Any} with 1 entry:
  "projection" => "PROJCS[\"WGS 84 / NSIDC EASE-Grid 2.0 Global\",GEOGCS[\"WGS 84\",DATUM[\"WGS_1984\",SP…
└────────────────────────────────────────────────────────────────────────────────────────────────────────┘
 ↓ →  6.0e6  5.99999e6  5.99998e6  5.99997e6  5.99996e6  …  5.91812e6  5.91811e6  5.9181e6  5.91809e6
 ⋮                                            ⋮          ⋱                                  

julia> @time pyr = PS.Pyramid(b1)
  1.303971 seconds (44.07 M allocations: 1.408 GiB, 4.77% gc time)
╭──────────────────────────────╮
│ 8192×8192 Pyramid{Float32,2} │
├──────────────────────────────┴─────────────────────────────────────────────────────────────────── dims ┐
  ↓ X Projected{Float64} LinRange{Float64}(400000.0, 481910.0, 8192) ForwardOrdered Regular Intervals{Start},
  → Y Projected{Float64} LinRange{Float64}(5.99999e6, 5.91808e6, 8192) ReverseOrdered Regular Intervals{Start}
├────────────────────────────────────────────────────────────────────────────────────────────── metadata ┤
  Metadata{Rasters.GDALsource} of Dict{String, Any} with 4 entries:
  "units"    => ""
  "offset"   => 0.0
  "filepath" => "test/data/crs_enabled_zarr_and_tif/some.tif"
  "scale"    => 1.0
└────────────────────────────────────────────────────────────────────────────────────────────────────────┘
 ⋮      ⋱  

Reproduce the Julia @time comparisons for PS.Pyramid(b1c) and PS.Pyramid(b1) using test/data/crs_enabled_zarr_and_tif/some.tif and the first-band examples in the issue. Trace the Pyramid entry points to identify the source of the timing difference; done should include a clear explanation and, if a change is warranted, benchmarked results.