Optimize memory usage by leveraging CommunityToolkit.HighPerformance for buffer pooling

[abnercpp]

This PR proposes leveraging the Microsoft official package CommunityToolkit.HighPerformance (alongside System.Memory and System.Collections.Immutable) to optimize memory usage by removing intermediate allocations where possible. This is only a small change to the project, but such is the nature of optimization: we can work with incremental steps.

Description

This PR aims to optimize memory usage for a few methods of the ImageMagick.MagickImage class, namely:

• void Trim(Gravity[] edges) now rents a buffer from the shared memory pool when converting edges to their string representations, which means we potentially get rid of intermediate allocations. (Also added a ReadOnlySpan<Gravity> overload for users that want to do their own pooling or stack allocate the parameters.)
• Methods writing to a stream or a file now potentially get rid of intermediate allocations caused by ToByteArray() by renting a buffer from the shared pool.
• ToByteArray() now potentially avoids intermediate allocations by writing the data into ArrayPoolBufferWriter<byte> instead of MemoryStream, which leverages the array pool to avoid unnecessary allocations, especially when dynamic resizes are needed.

[CLAassistant]

All committers have signed the CLA.

[dlemstra]

Thanks for creating this pull request but I don't want to add an extra dependency to this library for such a small change. But maybe we could use ArrayPool<byte>.Shared in the .NET Standard 2.1 code and reduce the allocations in WriteAsync? But I think we could also do that by creating a ByteArrayWrapper that uses an implementation that is similar to StreamWrapper to reduce allocations. I will see if I can make this change.

And to be honest the GravityExtensions really feels like an over optimization.

[dlemstra]

I am going to close this pull request because I don't want to add the change in this form.

[abnercpp]

Thanks for creating this pull request but I don't want to add an extra dependency to this library for such a small change. But maybe we could use ArrayPool<byte>.Shared in the .NET Standard 2.1 code and reduce the allocations in WriteAsync? But I think we could also do that by creating a ByteArrayWrapper that uses an implementation that is similar to StreamWrapper to reduce allocations. I will see if I can make this change.

And to be honest the GravityExtensions really feels like an over optimization.

Thank you for the fast and direct response. I really appreciate the feedback.

I can switch implementations by leveraging System.Memory directly. However that would require a dependency on System.Memory instead of CommunityToolkit.HighPerformance for netstandard2.0. (.NET6+ already comes with System.Memory.) Is that okay with you?

The gravity extensions were more of an attempt to clean up that part of the code than an optimization, but I can undo that with no problem.

[dlemstra]

I now have a proof of concept where I use the technique from the StreamWrapper in a new class to reduce the total allocated memory. Will need to make sure all my current test pass with this new class and add some extra tests before I commit these changes. This class will be used in the implementation of the ToByteArray method of the MagickImage class so I don't think we need extra improvements in that area. Maybe we could use an array pool there but because I need to change the size a lot I wonder if we should do this. We could have a look at it after my changes get merged. And thanks again for pointing this issue out.

I don't want to add an extra dependency to the netstandard2.0 code. But I am open for changes that improve parts of the netstandard2.1 code without adding extra dependencies.

p.s. If you are interested in improving performance and allocations then feel free to take a look at my magick-wasm project and see if I can do some smart things there. And I just realized that I will probably also need some help with using ToByteArray in the unit tests instead of writing and reading from a MemoryStream to reduce the allocations in the unit tests.