feat(buffer): add allocator-backed storage

[gatesn]

Summary

Add allocator-backed storage to vortex-buffer.

Design: vortex-data/rfcs#65

Changes

• Use allocator_api2 for owned buffers.
• Keep external buffers zero-copy.
• Preserve runtime alignment and bytes/Arrow interop.
• Add allocator-aware constructors and tests.

API Changes

Owned buffers no longer use bytes::Bytes as storage. The bytes crate remains for interop.

[codspeed-hq]

Merging this PR will regress 20 benchmarks

⚠️ 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

⚡ 81 improved benchmarks
❌ 20 regressed benchmarks
✅ 2110 untouched benchmarks
⏩ 64 skipped benchmarks¹
🗄️ 5 archived benchmarks run²

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	runend_compress_u32	398.3 µs	625.3 µs	-36.3%
❌	Simulation	non_nullable[32]	262.7 µs	398.5 µs	-34.08%
❌	Simulation	execute_scalar_struct_simple	111.5 µs	167.4 µs	-33.41%
❌	Simulation	non_nullable[2]	365.4 µs	537.6 µs	-32.04%
❌	Simulation	decompress[u32, (4000, 1024)]	55.7 µs	79.1 µs	-29.61%
❌	Simulation	nullable[2]	365.6 µs	499.3 µs	-26.78%
❌	Simulation	nullable[256]	259.3 µs	354.2 µs	-26.78%
❌	Simulation	nullable[32]	264.3 µs	360.5 µs	-26.69%
❌	WallTime	mul_u64_nonnull_neon	15.1 µs	20.3 µs	-25.69%
❌	Simulation	decompress[u8, (1000, 4)]	47 µs	63 µs	-25.41%
❌	Simulation	non_nullable[256]	258 µs	341 µs	-24.34%
❌	WallTime	lanezip_checked_add_u32_avx512[16384]	3.5 µs	4.5 µs	-22.45%
❌	Simulation	decode_varbin[(1000, 2)]	52.5 µs	66.2 µs	-20.77%
❌	WallTime	mul_i64_nonnull_neon	17.1 µs	20.7 µs	-17.64%
❌	WallTime	arrow_checked_add_u32_avx2[16384]	17.7 µs	21.4 µs	-17.03%
❌	WallTime	multiply_shapes_neon[(16384, PerRowPerRow)]	17.3 µs	20.7 µs	-16.47%
❌	Simulation	compress[(4000, 4)]	194.4 µs	232.2 µs	-16.25%
❌	Simulation	compress[(1000, 4)]	118.8 µs	140 µs	-15.13%
❌	Simulation	compress[(10000, 4)]	496 µs	575.4 µs	-13.8%
❌	Simulation	push_n_vortex_buffer[u32, 128]	2.8 µs	3.1 µs	-10.96%
...	...	...	...	...	...

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 ngates/buffer-allocator-core (719ac2e) with develop (01f147a)}

Footnotes

1. 64 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. ↩

2. 5 benchmarks were run, but are now archived. If they were deleted in another branch, consider rebasing to remove them from the report. Instead if they were added back, click here to restore them. ↩

[robert3005]

since I ran through experiments with this in the past - the Arc overhead here for small length dominates your allocations. tokio::Bytes plays game with pointer masking and stores AtomicPtr<()> and another pointer to avoid synchronistaion for trivial cases

[gatesn]

Yep... that's causing the slow downs I believe