Avoid unnecessary image decode.

[dcervelli]

image_source.js unnecessarily decodes image data via JavaScript before creating a Texture from the image data.

The Texture class can handle a native HTMLImageElement directly and then the browser can decode the image much more quickly and with less memory.

For my use case, this results in about a 10-20% reduction in scripting time.

[mourner]

It was originally like this but changed in this commit 2d359e9 — can you make sure this won't break the use cases that necessiated that change?

[dcervelli]

I pored over that commit and a few other related ones to determine what the intent was behind that change but don't have a definitive answer.

It seems like the change to image_source.js is unrelated to the broader commit (changing the public API of Map#addImage).

A couple of threads I tried to track down:

• Was the browser.getImageData call added to correspond to the desired change to the type in _prepareImage? This doesn't make a ton of sense though since the original HTMLImageElement should have been compatible with TexImageSource.
• Was the browser.getImageData call needed because most of the guts of dealing with texture were in image_source.js at the time (as opposed to texture.js, where they are now)? Again, this isn't an adequate explanation because the HTMLImageElement appears to be a sufficient type for the texture methods.

Given that image is private, the types are all compatible, and the changed line appears to be unrelated to the commit message where it was changed, this seems like a safe change.

@jfirebaugh, any background you could add?

[mourner]

I don't see why would it break anything too, so I think we should just merge — we can always revert later if anything comes up.