Values get clamped to typemax and typemin on write instead of `InexactError`

[tiemvanderdeure]

Values that cannot be represented by the eltype of a data silently clamped to typemin and typemax, where I would expect an InexactError as with other array types.

This might be intentional/documented behaviour? I ran into this through a Raster where that is not really obvious.

MWE

using ArchGDAL
out = "/tmp/jl_u5ihOBCFCl.tif"
width, height = 100, 100
data = fill(-1, width, height)  # (rows, cols)
driver = ArchGDAL.getdriver("GTiff")
ArchGDAL.create(out; driver, width, height, nbands = 1, dtype= UInt8) do ds
    band = ArchGDAL.getband(ds, 1)
    ArchGDAL.write!(band, data)
end

# similarly
ds = ArchGDAL.readraster(out; flags = ArchGDAL.OF_UPDATE)
ds[1] === 0x00
ds[1,2] = 1e10 # no error
ds[1,2] === typemax(UInt8) # value is clamped to max
close(ds)

[evetion]

Both your examples go through GDAL, and that does indeed clamp (https://gdal.org/en/stable/api/gdalrasterband_cpp.html). The question is whether we want to warn/error here?