opencl: fix two issues on flash attention for Adreno a7x (#25697)

* opencl: route `sub_group_shuffle_xor` to qcom ext when KHR ext is unavailable

KHR `sub_group_shuffle_xor` is not defined by compiler when
`cl_qcom_subgroup_shuffle` is present, causing certain FA
kernels fail to build. Define the KHR shuffle_xor using
the qcom extension.

* opencl: skip FA kernels with mixed and quant types for A7x to avoid compiler crash

ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl

@@ -30,6 +30,10 @@
 #elif defined(cl_qcom_subgroup_shuffle)
 #pragma OPENCL EXTENSION cl_qcom_subgroup_shuffle : enable
 #define HAS_SUBGROUP_SHUFFLE 1
+// Adreno compilers that expose only cl_qcom_subgroup_shuffle do not declare the KHR
+// name, so calling it is an implicit declaration and the program fails to build.
+// Route it to the qcom builtin.
+#define sub_group_shuffle_xor(val, mask) qcom_sub_group_shuffle_xor((val), (mask), CLK_SUB_GROUP_SHUFFLE_WIDTH_WAVE_SIZE_QCOM, 0.0f)
 #endif
 
 #define ACC_TYPE float