Unexpected table schema

It appears that Arrow.jl somehow incorrectly handles columns with nullable types (Union with Nothing) as the schema of the resulting table containing such a column doesn't match what's reported by `Tables.schema`.

For example, below is an example with a column of type `Union{Int64,Nothing}` that gets somehow interpreted by Arrow.jl as `Union{Missing, Nothing, Int64}` (should not include Missing).

```julia
julia> using Arrow

julia> tbl = Arrow.Table(Arrow.tobuffer((nothingints=Vector{Union{Int64,Missing}}([1,2,3,missing]),)))
Arrow.Table with 4 rows, 1 columns, and schema:
:nothingints Union{Missing, Int64}

julia> tbl = Arrow.Table(Arrow.tobuffer((nothingints=Vector{Union{Int64,Nothing}}([1,2,3,nothing]),)))
Arrow.Table with 4 rows, 1 columns, and schema:
:nothingints Union{Missing, Nothing, Int64}

julia> using Tables

julia> Tables.schema((nothingints=Vector{Union{Int64,Missing}}([1,2,3,missing]),))
Tables.Schema:
:nothingints Union{Missing, Int64}

julia> Tables.schema((nothingints=Vector{Union{Int64,Nothing}}([1,2,3,nothing]),))
Tables.Schema:
:nothingints Union{Nothing, Int64}
```

Reproduce the two Julia examples in the issue and compare Arrow.Table's reported schema with Tables.schema for columns typed as Union{Int64,Nothing}. Trace the schema handling for nullable columns; done means the Arrow table reports Union{Nothing,Int64} without adding Missing, while the existing Missing case remains correct.

Written by the indexing model from the issue text.