convert(Matrix, x') is very slow

convert(Matrix, x') is so slow that I haven't gotten it to finish. convert(Matrix, x) works just fine.

julia> using YAXArrays, Zarr

julia> path = "~/data/DataCube/v3.0.2/esdc-8d-0.25deg-1x720x1440-3.0.2.zarr"
"~/data/DataCube/v3.0.2/esdc-8d-0.25deg-1x720x1440-3.0.2.zarr"

julia> ds = open_dataset(path)
YAXArray Dataset
Shared Axes: 
↓ lon Sampled{Float64} -179.875:0.25:179.875 ForwardOrdered Regular Points,
→ lat Sampled{Float64} -89.875:0.25:89.875 ForwardOrdered Regular Points,
↗ Ti  Sampled{DateTime} [1979-01-05T00:00:00, …, 2021-12-31T00:00:00] ForwardOrdered Irregular Points
Variables: 
....


julia> mat = ds.air_temperature_2m[Ti = 1000]
╭──────────────────────────────╮
│ 1440×720 YAXArray{Float32,2} │
├──────────────────────────────┴──────────────────────────────────────── dims ┐
  ↓ lon Sampled{Float64} -179.875:0.25:179.875 ForwardOrdered Regular Points,
  → lat Sampled{Float64} -89.875:0.25:89.875 ForwardOrdered Regular Points
├──────────────────────────────────────────────────────
....

julia> @time convert(Matrix{Float32}, mat)
  1.809297 seconds (5.15 M allocations: 362.756 MiB, 5.92% gc time, 98.13% compilation time)
...

julia> @time convert(Matrix{Float32}, mat')

# has to be interrupted

Start by reproducing the Julia example with YAXArrays and Zarr, comparing convert(Matrix{Float32}, mat) with convert(Matrix{Float32}, mat'). Trace the conversion path for the transposed YAXArray and verify that the transposed conversion completes without the severe slowdown or excessive allocations shown.