opencl: add q6_K gemm and gemv kernels for Adreno

[lhez]

This PR adds Q6_K gemm and gemv kernels for Adreno. This should improve performance for models containing Q6_K quantization.

For Q4_K_M, we will need to go through the same for Q4_K. Therefore, Q4_K_M is still slow but should be better.

On X Elite,

before,

Qwen3-0.6B-Q6_K,

common_perf_print: prompt eval time =     724.44 ms /   235 tokens (    3.08 ms per token,   324.39 tokens per second)
common_perf_print:        eval time =   13167.73 ms /   256 runs   (   51.44 ms per token,    19.44 tokens per second)

Qwen3-4B-Q6_K,

common_perf_print: prompt eval time =    4901.20 ms /   235 tokens (   20.86 ms per token,    47.95 tokens per second)
common_perf_print:        eval time =   51144.81 ms /   256 runs   (  199.78 ms per token,     5.01 tokens per second)

Qwen3-0.6B-Q4_K_M,

common_perf_print: prompt eval time =    1514.15 ms /   235 tokens (    6.44 ms per token,   155.20 tokens per second)
common_perf_print:        eval time =    8231.90 ms /   256 runs   (   32.16 ms per token,    31.10 tokens per second)

Qwen3-4B-Q4_K_M.gguf,

common_perf_print: prompt eval time =   11502.19 ms /   235 tokens (   48.95 ms per token,    20.43 tokens per second)
common_perf_print:        eval time =   28561.40 ms /   256 runs   (  111.57 ms per token,     8.96 tokens per second)

after,

Qwen3-0.6B-Q6_K,

common_perf_print: prompt eval time =     281.45 ms /   235 tokens (    1.20 ms per token,   834.95 tokens per second)
common_perf_print:        eval time =    4243.57 ms /   256 runs   (   16.58 ms per token,    60.33 tokens per second)

Qwen3-4B-Q6_K,

common_perf_print: prompt eval time =    1605.23 ms /   235 tokens (    6.83 ms per token,   146.40 tokens per second)
common_perf_print:        eval time =   23625.54 ms /   256 runs   (   92.29 ms per token,    10.84 tokens per second)

Qwen3-0.6B-Q4_K_M,

common_perf_print: prompt eval time =    1497.88 ms /   235 tokens (    6.37 ms per token,   156.89 tokens per second)
common_perf_print:        eval time =    5191.81 ms /   256 runs   (   20.28 ms per token,    49.31 tokens per second)

Qwen3-4B-Q4_K_M.gguf,

common_perf_print: prompt eval time =   10775.26 ms /   235 tokens (   45.85 ms per token,    21.81 tokens per second)
common_perf_print:        eval time =   25510.26 ms /   256 runs   (   99.65 ms per token,    10.04 tokens per second)