feat[gpu]: slice support for dyn dispatch

[0ax1]

Slice offsets are either applied during the dyn dispatch build plan construction or in case of bitpacking within the dyn dispatch kernel, as sub-byte offsets can't be expressed by pointer arithmetic on the device ptr.

Calling output_tile_len as part of the final execute_stage loop has no perf overhead compared to develop.

[codspeed-hq]

Merging this PR will improve performance by 12.36%

⚡ 1 improved benchmark
✅ 999 untouched benchmarks
⏩ 1466 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	bitwise_not_vortex_buffer_mut[128]	530.3 ns	471.9 ns	+12.36%

---

_{Comparing ad/gpu-dyn-slice (cab2c97) with develop (aff52ad)}

Footnotes

1. 1466 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. ↩

[a10y]

Why adding a new method vs fixing cuda_device_ptr()? We use cuda_device_ptr in a number of places where we probably mean offset_ptr()

[a10y]

Oh I see, cuda_device_ptr becomes an alias for offset_ptr now