spike: chunked execution engine for primitive + listview decode

[joseph-isaacs]

Introduces a streaming chunked decode engine alongside the canonical
executor. The model is:

• Driver-owned L1-resident Scratch of fixed CHUNK_LEN (1024); producers
  write decoded values into it and the driver consumes per chunk.
• PrimitiveChunkProducer contract + dyn-dispatch
  PrimitiveChunkKernelDispatcher keyed by outermost encoding id.
• DictKernel materializes its (bounded) values slot once via the regular
  executor, then streams the gather over codes; this naturally fuses
  Dict<RunEnd

  > since RunEnd is unrolled into the small dict in the
  materialization step.

• RunEndKernel lives in vortex-runend (where the encoding is defined) and
  registers onto the dispatcher via register_chunk_kernels.
• listview::ListChunkProducer emits (offsets, sizes, elements) row windows
  for ListView, including bit-packed offsets/sizes.

Module is _-prefixed (#[doc(hidden)]) so it stays out of the public API
surface while the spike settles.

Includes:

• unit tests in vortex-array (slice round-trip, dict chunked, listview
  windows, fallback) and vortex-runend (runend, sliced runend, fused
  Dict<RunEnd

  >).

• divan bench at encodings/runend/benches/chunked_exec.rs comparing
  chunked vs canonical for Dict

  , RunEnd

  , fused Dict<RunEnd

  >,
  and ListView

  with bit-packed offsets/sizes.

Signed-off-by: Claude claude@anthropic.com

[codspeed-hq]

Merging this PR will degrade performance by 13.16%

⚠️ 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.

❌ 2 regressed benchmarks
✅ 1235 untouched benchmarks
🆕 98 new benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	cuda/bitpacked_u8/unpack/3bw[100M]	298.9 µs	351.6 µs	-14.98%
❌	Simulation	chunked_varbinview_canonical_into[(100, 100)]	273.1 µs	308 µs	-11.31%
🆕	Simulation	dict_runend_canonical[len=4194304 dict=1024 inner_run=8]	N/A	17.9 ms	N/A
🆕	Simulation	dict_runend_fused_chunked[len=1048576 dict=4096 inner_run=16]	N/A	4.8 ms	N/A
🆕	Simulation	dict_primitive_chunked[len=262144 dict=1024]	N/A	1.2 ms	N/A
🆕	Simulation	dict_runend_canonical[len=1048576 dict=256 inner_run=4]	N/A	4.5 ms	N/A
🆕	Simulation	dict_runend_fused_chunked[len=1048576 dict=256 inner_run=4]	N/A	4.7 ms	N/A
🆕	Simulation	dict_primitive_chunked[len=16777216 dict=256]	N/A	74.5 ms	N/A
🆕	Simulation	dict_primitive_chunked[len=65536 dict=256]	N/A	300.1 µs	N/A
🆕	Simulation	dict_runend_phase_inner_canonical[len=1048576 dict=256 inner_run=4]	N/A	16.1 µs	N/A
🆕	Simulation	dict_bp_canonical[len=1048576 dict=1024 bw=10]	N/A	4.3 ms	N/A
🆕	Simulation	dict_primitive_chunked[len=4194304 dict=256]	N/A	18.6 ms	N/A
🆕	Simulation	dict_runend_canonical[len=1048576 dict=4096 inner_run=16]	N/A	4.6 ms	N/A
🆕	Simulation	dict_bp_canonical[len=1048576 dict=256 bw=8]	N/A	4.1 ms	N/A
🆕	Simulation	dict_bp_canonical[len=4194304 dict=256 bw=8]	N/A	20.1 ms	N/A
🆕	Simulation	dict_bp_canonical[len=1048576 dict=4096 bw=12]	N/A	4.5 ms	N/A
🆕	Simulation	dict_runend_phase_inner_canonical[len=1048576 dict=4096 inner_run=16]	N/A	30.7 µs	N/A
🆕	Simulation	dict_runend_phase_inner_canonical[len=4194304 dict=1024 inner_run=8]	N/A	20.2 µs	N/A
🆕	Simulation	dict_runend_phase_take[len=4194304 dict=1024 inner_run=8]	N/A	5.3 µs	N/A
🆕	Simulation	listview_canonical_sum[rows=65536 avg_list=4 bw_off=18 bw_sz=4]	N/A	1.8 ms	N/A
...	...	...	...	...	...

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 claude/vortex-execution-engine-EUlmE (bcf5b76) with develop (19a1fb3)}