fix cuda_device_ptr() method

[a10y]

Summary

Previously we'd return the pointer to the front of the allocation.

We should be returning the pointer to the beginning of the pointed range, e.g. for sliced allocations like what happens when you try and load the patches out of a CudaFlatLayout.

I encountered this after wasting a lot of time trying to unpack why patches.indices.buffer_handle() and patches.values.buffer_handle() were the same pointer.

Testing

Added a simple unit test + existing test suite.

[a10y]

Github is throwing another one of its tantrums it seems

[codspeed-hq]

Merging this PR will improve performance by 12.36%

⚡ 2 improved benchmarks
✅ 998 untouched benchmarks
⏩ 1466 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	map_each[BufferMut<i32>, 128]	858.1 ns	770.6 ns	+11.36%
⚡	Simulation	bitwise_not_vortex_buffer_mut[128]	530.3 ns	471.9 ns	+12.36%

---

_{Comparing aduffy/device-ptr-fix (0eb716e) with develop (6ff018b)}

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. ↩

[robert3005]

I think #6862 might already fix it