Remove indexmap

[nevi-me]

Which issue does this PR close?

Closes #1882.

Rationale for this change

See #1882

What changes are included in this PR?

Removes the indexmap dependency to evaluate the impact on tests

Are there any user-facing changes?

Yes, arrow::json::reader::ReaderBuilder::with_format_strings used to incorrectly take an IndexMap aliased as HashMap, it and other related decoder options now take a HashMap

---

Note that there are other crates that depend on indexmap.

These are:

• clap - integration-testing, in parquet as an optional feature for binaries
• h2 - arrow-flight (dependency of tonic)
• petgraph - arrow-flight arrow-flight (dependency of tonic-build)
• tower - arrow-flight (dependency of tonic)

So if we can find a suitable resolution to the JSON writer, we could remove indexmap from the arrow crate.

I've also noticed that our write behaviour changes if we don't preserve_order on serde-json, which could be a blocker.

[nevi-me]

@alamb @jhorstmann @tustvold please see comments on the PR.
I haven't looked at the json writer failures, but I know that they depend on serde_json's preserve_order feature. So there might not be much of an alternative there.

One approach could be to isolate the json functionality into its own feature, and perhaps even go as far as splitting the read and write features to be under the json feature. It would at least enable us to remove indexmap if one doesn't need json_write