switch Buffer -> BufferHandle in PrimitiveArray

[a10y]

Fleshes out the BufferHandle type more.

I hid the inner enum and introduced some extra methods to build/unwrap BufferHandles.

This removes the PrimitiveArray::as_slice method, replacing it with an into_buffer::<T> method which may allocate and copy a new buffer (if the handle points to device memory), or else provides a new Buffer pointing to the existing host memory.

This PR is large but a lot of it is just adding &'s and updating test code to use the buffer! macro

[codspeed-hq]

Merging this PR will degrade performance by 66.29%

❌ 42 regressed benchmarks
✅ 389 untouched benchmarks
⏩ 823 skipped benchmarks¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	decompress[u8, (1000, 256)]	8.6 µs	9.7 µs	-11.18%
❌	decompress[u16, (1000, 256)]	9.9 µs	11 µs	-10.63%
❌	decompress_alp[f32, (1000, 0.0, 0.95)]	9.3 µs	13.3 µs	-29.78%
❌	decompress_alp[f32, (1000, 0.0, 1.0)]	9.2 µs	13.1 µs	-30.1%
❌	decompress_alp[f32, (1000, 0.01, 0.95)]	12.5 µs	17.5 µs	-28.18%
❌	decompress_alp[f32, (10000, 0.01, 0.95)]	32.4 µs	88.4 µs	-63.39%
❌	decompress_alp[f32, (1000, 0.01, 0.25)]	12.7 µs	18.1 µs	-29.72%
❌	decompress_alp[f32, (1000, 0.01, 1.0)]	12.4 µs	17.9 µs	-31.05%
❌	decompress_alp[f32, (10000, 0.1, 0.25)]	35.1 µs	90.6 µs	-61.27%
❌	decompress_alp[f32, (1000, 0.1, 1.0)]	13.4 µs	19.1 µs	-30.18%
❌	decompress_alp[f32, (10000, 0.0, 0.95)]	27.9 µs	82.7 µs	-66.23%
❌	decompress_alp[f32, (1000, 0.1, 0.95)]	14 µs	18.8 µs	-25.45%
❌	decompress_alp[f32, (1000, 0.1, 0.25)]	12.6 µs	17.8 µs	-29.27%
❌	decompress_alp[f32, (10000, 0.0, 1.0)]	27.8 µs	82.4 µs	-66.27%
❌	decompress_alp[f32, (10000, 0.0, 0.25)]	27.9 µs	82.8 µs	-66.24%
❌	decompress_alp[f32, (10000, 0.1, 0.95)]	42.2 µs	97.7 µs	-56.83%
❌	decompress_alp[f32, (10000, 0.01, 0.25)]	31.7 µs	87.8 µs	-63.89%
❌	decompress_alp[f64, (10000, 0.01, 0.95)]	59.3 µs	167.8 µs	-64.66%
❌	decompress_alp[f32, (10000, 0.01, 1.0)]	32.9 µs	88.6 µs	-62.83%
❌	decompress_alp[f64, (10000, 0.0, 1.0)]	54.6 µs	161.9 µs	-66.29%
...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

---

_{Comparing bufferhandles (50da3c2) with develop (2b2ad23)}

Footnotes

1. 823 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. ↩

[joseph-isaacs]

I wonder if we can remove at lot of these buffer calls from tests?