EXRLoader: support single channel luminance - RedFormat

[sciecode]

Addresses #20195 (comment)

I noticed a difference between our implementation and OpenEXR's in trying to decompress a block only if input-data is actually smaller than the expected data-size. This PR standardizes this issue and makes equivalent to OpenEXR implementation.

It also initially includes support for luminance - single channel files HalfFloatType / FloatType.
Luminance images are treated as RedFormat - WebGL2 only context.

Updated textureData header info to include specification flags, to help identify if problematic files contain lack of supported features like Deep-Image, Multi-Part or Tiled image files.

I largely tested these changes, but would appreciate further testing - @Mugen87, @WestLangley

--

Currently we fail to support .setDataType( THREE.UnsignedByteType ) on Luminance maps.

setDataType is properly followed even in cases where there is an underlying mismatch between input-type and output-types.
( e.g like a float32b input file transformed into a half16b texture ). When we set the output-type to UnsignedByteType on a RGB/RGBA files we also encode the output-texture as RGBEEncoding and transform the data appropriately.

However transforming a luminance map of 32/16 type to an uint8 type is not yet specified, should I explicitly allow that transformation? If so, should I then change the texture format to LuminanceFormat in order to support WebGL1 contexts?

[Mugen87]

However transforming a luminance map of 32/16 type to an uint8 type is not yet specified, should I explicitly allow that transformation?

I think at some point the encoding to RGBE should be removed from EXRLoader and half float used instead. With WebGL 2 we can rely on half float, see #22998.

Yes, half float textures allocate more memory however it's possible to use THREE.LinearFilter and it simplifies the code in the loader and shader.

[Mugen87]

The luminance texture on my computer loads now as expected 👍 .

[sciecode]

However transforming a luminance map of 32/16 type to an uint8 type is not yet specified, should I explicitly allow that transformation?

I think at some point the encoding to RGBE should be removed from EXRLoader and half float used instead. With WebGL 2 we can rely on half float, see #22998.

Loader already specifies HalfFloatType as the default output type.

Yes, half float textures allocate more memory however it's possible to use THREE.LinearFilter and it simplifies the code in the loader and shader.

Yeah I can definitely see that being an issue. In that case, this PR should be ready to merge, in case we don't find any testing problems.

[mrdoob]

Thanks!