4x Faster BlurHash decoding with optimized algorithm

[Rahiche]

Key Changes

Added a new BlurHashOptimizationMode enum with three options:

1. none: Uses the original algorithm (default to be backward compatible)
2. standard: Optimized with better cache locality (1.9)
3. approximation: Fast approximation with simplified sRGB conversion (a slightly darker tint)

standard mode does the following:

• Improved cache locality
• Pre-calculated cosine values to avoid redundant computations

approximation mode add this:

• faster approximation method for sRGB conversion

Migration

This change is fully backward compatible. The default optimizationMode for BlurHashWidget is set to none, so existing code will continue to use the original algorithm. To take advantage of the optimizations, users can set the optimizationMode parameter to standard or approximation.

Mode Comparison Screenshots

Standard Mode (Release Mode) (1.9x faster)

Standard Mode

Size	Screenshot
512×512
256×256
128×128
64×64

Approximation Mode (Release Mode) (4x faster)

Size	Screenshot
512×512
256×256
128×128
64×64

[mark8044]

@Rahiche Standard mode is working very well, I can't even notice a difference from default. very nice.

However approximation mode creates a weird pixelated shape

None/Standard:

Approximation:

Blurhash i was using for this picture: A44.JD_48_4n

This is happening for any blur hash I throw at it

[Rahiche]

Thanks for catching this @mark8044
if you can create an issue I'll make a fix ASAP

[fingerart]

@Rahiche Where is the source code of benchmarks?