test[gpu]: more e2e tests

[0ax1]

No description provided.

[codspeed-hq]

Merging this PR will not alter performance

⚠️ 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.

⚡ 4 improved benchmarks
❌ 5 regressed benchmarks
✅ 1510 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	chunked_bool_canonical_into[(1000, 10)]	32 µs	46.9 µs	-31.74%
❌	Simulation	chunked_varbinview_canonical_into[(1000, 10)]	162.4 µs	198.6 µs	-18.22%
❌	Simulation	chunked_varbinview_into_canonical[(1000, 10)]	177.2 µs	213.5 µs	-17%
❌	Simulation	decompress_rd[f32, (100000, 0.0)]	499.1 µs	586.5 µs	-14.9%
❌	Simulation	decompress_rd[f64, (100000, 0.0)]	845.5 µs	981.8 µs	-13.88%
⚡	WallTime	cuda/bitpacked_u8/unpack/3bw[100M]	352.3 µs	299.1 µs	+17.79%
⚡	Simulation	bitwise_not_vortex_buffer_mut[128]	275.3 ns	246.1 ns	+11.85%
⚡	Simulation	decompress_rd[f64, (100000, 0.01)]	981.3 µs	889.4 µs	+10.33%
⚡	Simulation	decompress_rd[f64, (100000, 0.1)]	981.2 µs	889.4 µs	+10.33%

Tip

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

---

_{Comparing ad/e2e (07b233f) with develop (3e68153)}