Possible bug in `Any` concretization routine

As discussed on Slack:

```
julia> using Arrow, DataFrames

julia> Arrow.write("test.arrow", (a = [1, 2], b = Any[3, 4.5]))
"test.arrow"

julia> DataFrame(Arrow.Table("test.arrow"))
2×2 DataFrame
Row │ a b
│ Int64 Float64
─────┼─────────────────
1 │ 1 1.5e-323
2 │ 2 4.5

(jl_aYpToJ) pkg> st
Status `C:\Users\ngudat\AppData\Local\Temp\jl_aYpToJ\Project.toml`
[69666777] Arrow v2.2.0
```

Reproduce the example in Julia with Arrow and DataFrames, using the shown Arrow.write and Arrow.Table calls. Then trace the Any concretization routine involved in reading the mixed-integer-and-float column; done means the resulting DataFrame preserves the values 3 and 4.5 instead of producing 1.5e-323 for the first value.

Written by the indexing model from the issue text.