Underlying file still referenced even with `copycols=true`

Hello,

I am getting errors when trying to delete a file loaded with `DataFrame(Arrow.Table(…))`.
It seems that the underlying file is still referenced even when I do it with `copycols=true`.

```
n = 1
filename = "$(randstring(12)).arrow"
df = DataFrame(x=rand(n))
Arrow.write("$filename", df)
res1 = DataFrame(Arrow.Table("$filename"); copycols=true)
rm(filename)
```

This the output:

```
PS > julia -q --project .\test.jl
ERROR: LoadError: IOError: unlink("iPJQf2OSiQaT.arrow"): permission denied (EACCES)
Stacktrace:
[1] uv_error
@ .\libuv.jl:97 [inlined]
[2] unlink(p::String)
@ Base.Filesystem .\file.jl:934
[3] rm(path::String; force::Bool, recursive::Bool)
@ Base.Filesystem .\file.jl:272
[4] rm(path::String)
@ Base.Filesystem .\file.jl:263
[5] top-level scope
@ D:\source\COMPILETHIS\test.jl:13
in expression starting at test.jl:13
```

This happens with Julia 1.6.1, Arrow v1.5.0, Windows 10

Run the supplied Julia reproduction on Windows 10, starting with DataFrame(Arrow.Table("$filename"); copycols=true) and the subsequent rm(filename). Trace how the Arrow.Table input is retained during DataFrame construction; done means the file can be deleted successfully after construction with copycols=true.

Written by the indexing model from the issue text.