feat(ext/canvas): enhance createImageBitmap specification compliance

[Hajime-san]

Feature

basic support for decoding image format

• jpeg, png, gif, bmp, ico, webp
  • animated image
  • 16-bit image
  • Grayscale image

support colorSpace for ImageData

There can convert color space from srgb to display-p3 with createImageBitmap.

This was a misunderstanding on my part, so I reverted it.
The conversion using the ImageData colorSpace option was probably the responsibility of CanvasRenderingContext2D.
I've left the implementation commented out as it may be useful as a reference for implementing CanvasRenderingContext2D or OffscreenCanvasRenderingContext2D.

support colorSpaceConversion option

The specifications are not particularly clear about the implementation details, and I had a hard time understanding its behavior, but I eventually do reverse engineering the results of wpt and implemented it to be consistent with those results.

• createImageBitmap: Clarify treatment of colorSpaceConversion:"none" ImageBitmaps whatwg/html#10578

port to Rust

To simplify code, almost logic moved to Rust side.

added a simple architecture document for README

It is basically based on the image implementation, and I wrote roughly how you can handle data effectively.

performance impact

I was curious to see if the increased amount of Rust code, especially static dispatch, was affecting the bootstrap time of the runtime.
I'm not very confident about the suitability of this benchmark, but it may not have had a big impact.

Deno.bench("bootstrap time", () => {
  null
})
Deno.bench("createImageBitmap", async () => {
  const imageData = new ImageData(new Uint8ClampedArray([255, 0, 0, 255]), 1, 1);
  await createImageBitmap(imageData);
})

• this pr 86f0000

% target/release/deno bench bench.ts
Check file:///path/to/deno/bench.ts
    CPU | Apple M1
Runtime | Deno 2.0.0-rc.1 (aarch64-apple-darwin)

file:///path/to/deno/bench.ts

benchmark           time/iter (avg)        iter/s      (min … max)           p75      p99     p995
------------------- ----------------------------- --------------------- --------------------------
bootstrap time               4.3 ns   232,900,000 (  3.9 ns …   1.0 µs)   4.1 ns   7.3 ns  10.2 ns
createImageBitmap            6.8 µs       146,300 (  5.0 µs …  26.4 ms)   5.6 µs  19.8 µs  26.4 µs

• main b1b72a8

% target/release/deno bench bench.ts  
Check file:///path/to/deno/bench.ts
    CPU | Apple M1
Runtime | Deno 1.46.1 (aarch64-apple-darwin)

file:///path/to/deno/bench.ts

benchmark           time/iter (avg)        iter/s      (min … max)           p75      p99     p995
------------------- ----------------------------- --------------------- --------------------------
bootstrap time               4.9 ns   205,300,000 (  4.2 ns … 101.1 ns)   4.3 ns  11.0 ns  13.5 ns
createImageBitmap            4.6 µs       218,000 (  4.4 µs …   5.0 µs)   4.6 µs   5.0 µs   5.0 µs

plans for enhancement in another PRs

• improve performance
  There is known to that processing image is a cpu-bound task. So, op_create_image_bitmap will be async fn and consider to be data-parallelism with using rayon.

• align error message

  • Normalize error message grammar (TS/JS) #25269
  • refactor(ext): align error messages #25310