Arrow changes data type from input in unexpected ways

From this MWE input is unrecognizable from output (the path to the Zarr file is public and can be run locally):

```
dc = Zarr.zopen("http://its-live-data.s3.amazonaws.com/datacubes/v02/N20E100/ITS_LIVE_vel_EPSG32647_G0120_X650000_Y3250000.zarr")
C = dc["satellite_img1"][:]
input = DataFrame([C,C],:auto)
Arrow.write("test.arrow", input)
output = Arrow.Table("test.arrow")
```

`input.x1` looks like this:
```
1460-element Vector{Zarr.MaxLengthStrings.MaxLengthString{2, UInt32}}:
"1A"
⋮
"8."
```

while `output.x1` looks like this:
```
1460-element Arrow.List{String, Int32, Vector{UInt8}}:
"1\0"
⋮
"\0\0"
```

Start with the linked MWE using Arrow.write and Arrow.Table, then compare the input.x1 and output.x1 types and values for the Zarr.MaxLengthString data. Determine the intended round-trip representation and verify that the serialized output preserves the input values and expected type.

Written by the indexing model from the issue text.