edd4d9bca5
Add dequantize4() implementations for Q4_1, Q5_0, Q5_1, and IQ4_NL in the flash attention base shader. Register them in the shader generator, pipeline creation, and enable in the scalar/coopmat1 FA support check.