feat(vortex-geo): ST_Distance via the geo crate + Polygon type

[HarukiMoriarty]

Summary

1. offload the math to geo: unification, one distance (and remaining geo functions) implementation, replace the hand-roll kernel.
2. support polygon: as a second native type riding that same path
3. all geometry decode goes through Arrow/geoarrow

The performance decreased (5.6ms -> 12ms on Q1) compared with the old direct-coordinate kernel, it is a tradeoff for one unified, general decode path. The cost is the intermediate allocation/enumeration, not the math itself. Type-specialized fast paths (e.g. point-point straight off the coordinate struct) can recover it later without changing the API.

Testing

4 distance behavior tests through geo crate, and tests for polygon type.

[codspeed-hq]

Merging this PR will improve performance by 22.57%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 93 improved benchmarks
❌ 12 regressed benchmarks
✅ 1476 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	take_10k_random	197.7 µs	253.7 µs	-22.08%
❌	Simulation	take_10k_contiguous	217.6 µs	274.4 µs	-20.68%
❌	Simulation	patched_take_10k_contiguous_patches	232.1 µs	288.8 µs	-19.64%
❌	Simulation	patched_take_10k_random	244.5 µs	301.2 µs	-18.83%
❌	Simulation	encode_varbin[(1000, 2)]	239.6 µs	294.6 µs	-18.67%
❌	Simulation	chunked_varbinview_canonical_into[(1000, 10)]	162.7 µs	191.4 µs	-14.98%
❌	Simulation	varbinview_large	112.6 µs	130.8 µs	-13.93%
❌	Simulation	decompress_rd[f64, (100000, 0.01)]	890.1 µs	1,020.4 µs	-12.77%
❌	Simulation	decompress_rd[f64, (100000, 0.1)]	890.2 µs	1,020.4 µs	-12.76%
❌	Simulation	and_true_constant	14.9 µs	16.8 µs	-11.41%
❌	Simulation	or_false_constant	14.9 µs	16.8 µs	-11.23%
❌	Simulation	bench_many_codes_few_values[1024]	467.5 µs	526.3 µs	-11.17%
⚡	Simulation	search_index_mixed_out_of_range	255.9 µs	78.9 µs	×3.2
⚡	Simulation	search_index_above_max	255.7 µs	78.9 µs	×3.2
⚡	Simulation	search_index_below_min	255.7 µs	78.9 µs	×3.2
⚡	Simulation	search_index_full_range_random	255.9 µs	79 µs	×3.2
⚡	Simulation	search_index_in_range	256.1 µs	79.2 µs	×3.2
⚡	Simulation	chunked_dict_primitive_into_canonical[u32, (1000, 10, 10)]	185.4 µs	87.3 µs	×2.1
⚡	Simulation	chunked_bool_canonical_into[(10, 1000)]	751.4 µs	492.6 µs	+52.55%
⚡	Simulation	chunked_opt_bool_canonical_into[(10, 1000)]	866.6 µs	572.3 µs	+51.42%
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing vortex-geo-pr (242ee4c) with develop (aef6307)¹}

Footnotes

1. No successful run was found on develop (ed69077) during the generation of this report, so aef6307 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩