GH-40053: [Python] Preserve dict key order when inferring struct type

[gabrielsimoes]

Rationale for this change

Fixes #40053

When converting Python dictionaries to PyArrow arrays, struct fields are sorted alphabetically instead of preserving the original dictionary key insertion order. Since Python 3.7+, dictionaries maintain insertion order, and users expect this order to be preserved.

>>> import pyarrow as pa
>>> pa.array([{"b": 2, "a": 1}]).type
struct<a: int64, b: int64>

Expected: struct<b: int64, a: int64>

What changes are included in this PR?

Replace std::map<std::string, TypeInferrer> with std::vector<std::pair<std::string, TypeInferrer>> + std::unordered_map<std::string, size_t> in the type inference code. This follows the same pattern used in the JSON parser (cpp/src/arrow/json/parser.cc) for the same problem.

Are these changes tested?

Updated existing tests to verify field ordering.

Are there any user-facing changes?

Struct field order now matches dictionary key insertion order instead of being sorted alphabetically. This is a behavioral change but aligns with user expectations and Python semantics.

• GitHub Issue: [Python] pa.array(<pd.Series of structs>) changes field order to be sorted #40053

[github-actions]

⚠️ GitHub issue #40053 has been automatically assigned in GitHub to PR creator.

[github-actions]

⚠️ GitHub issue #40053 has been automatically assigned in GitHub to PR creator.

[pitrou]

Thanks for the PR @gabrielsimoes ! The solution looks good in general, here are some relatively minor comments.

[gabrielsimoes]

Thanks for the speedy review @pitrou; I have addressed them and the linter/doctest CI failures.

[github-actions]

⚠️ GitHub issue #40053 has been automatically assigned in GitHub to PR creator.

[pitrou]

This looks ok to me, thank you @gabrielsimoes . I'll wait for CI to pass and then I think we can merge.

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 3 benchmarking runs that have been run so far on merge-commit cff2c52.

There weren't enough matching historic benchmark results to make a call on whether there were regressions.

The full Conbench report has more details.