[CPU]Fix by_channel quant for avx2 (#29532)

### Details:
 - *Fix dot_product kernel with avx2 support*

### Tickets:
 - *ticket-id*

Author: zhangYiIntel

src/plugins/intel_cpu/src/nodes/kernels/scaled_attn/attn_quant.cpp

@@ -275,8 +275,8 @@ static void quant_u8_by_channel_kernel(const T* src,
         }
     }
 #endif
-    for (size_t i = 0; i < seq_dim; ++i) {
-        for (; j < hidden_dims; j++) {
+    for (; j < hidden_dims; j++) {
+        for (size_t i = 0; i < seq_dim; ++i) {
             float tmp = src[i * src_stride + j];
             dst[i * dst_stride + j] = static_cast<uint8_t>(std::round(tmp / scale[j] + zp[j]));
         }

src/plugins/intel_cpu/src/nodes/kernels/scaled_attn/executor_pa.cpp

@@ -681,10 +681,10 @@ static void dot_product_block_by_channel(TA* a, uint8_t* b, float* c, const size
             auto va2 = mm256_uni_loadu_ps(a + i + vec_len_f32_avx2 * 2);
             auto va3 = mm256_uni_loadu_ps(a + i + vec_len_f32_avx2 * 3);
 
-            auto vb0_128 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(b + i));
-            auto vb1_128 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(b + i + vec_len_f32_avx2));
-            auto vb2_128 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(b + i + vec_len_f32_avx2 * 2));
-            auto vb3_128 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(b + i + vec_len_f32_avx2 * 3));
+            auto vb0_128 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(b + params_offset + i));
+            auto vb1_128 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(b + params_offset + i + vec_len_f32_avx2));
+            auto vb2_128 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(b + params_offset + i + vec_len_f32_avx2 * 2));
+            auto vb3_128 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(b + params_offset + i + vec_len_f32_avx2 * 3));
 
             auto vb0_256 = _mm256_cvtepu8_epi32(vb0_128);
             auto vb1_256 = _mm256_cvtepu8_epi32(vb1_128);