pickling support

[onursatici]

Implements __reduce__ and __reduce_ex__ methods to enable pickling of Vortex arrays in Python. Arrays are serialized using the Vortex IPC format.
For pickle protocol 5+ (Python 3.8+, PEP 574), uses PickleBuffer to keep array buffers separate from the main pickle stream rather than copying them inline. This enables us to use shared memory in the future to potentially zero-copy large arrays even across process boundaries. Protocol 4 and below serialise buffers inline as bytes.
Both protocols share the same deserialization path via decode_ipc_array_buffers, which reconstructs arrays from IPC-encoded buffer lists (or memoryviews).

[onursatici]

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 84.87%. Comparing base (28f5b3d) to head (62f5e86).
⚠️ Report is 23 commits behind head on develop.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[onursatici]

better

[gatesn]

I think you should expose this via the main vortex crate

[gatesn]

You're gonna need a PR description for this one 😅

[gatesn]

I don't understand the two screenshots you sent? Also, is there any point implementing P4 at all? We're way beyond 3.8 as a min version requirement.

[onursatici]

🤦‍♂️ right. I did initially add 4 to be a baseline to see if 5 would make a difference. Screenshots show the benchmarks ran comparing 4 vs 5. You are right we have min 3.11 required for vortex python so v4 would not be used, I will remove

[onursatici]

@gatesn protocol 5 is introduced on py3.8, but made default just recently at py3.14, so we still need both

[codspeed-hq]

CodSpeed Performance Report

Merging #5011 will not alter performance

_{Comparing os/pickle (62f5e86) with develop (ee4b921)¹}

Summary

✅ 1318 untouched
🆕 7 new
⏩ 162 skipped²

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
🆕	decompress[("alp_for_bp_f64", 0x75e2f0)]	N/A	24.2 ms	N/A
🆕	decompress[("datetime_for_bp", 0x7615f0)]	N/A	34.9 ms	N/A
🆕	decompress[("dict_fsst_varbin_bp_string", 0x760700)]	N/A	14.5 ms	N/A
🆕	decompress[("dict_fsst_varbin_string", 0x760070)]	N/A	14.5 ms	N/A
🆕	decompress[("dict_varbinview_string", 0x75ed10)]	N/A	14.7 ms	N/A
🆕	decompress[("for_bp_u64", 0x75dba0)]	N/A	2.5 ms	N/A
🆕	decompress[("runend_for_bp_u32", 0x75f1c0)]	N/A	2 ms	N/A

Footnotes

1. No successful run was found on develop (99101b1) during the generation of this report, so ee4b921 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

2. 162 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩