[CPU] oneDNN3.6 cherry-pick backports (openvinotoolkit#28843)

- fix the copy_a init in brgemm:
uxlfoundation/oneDNN@59513f7
- relax the rnn tests

test onednn branch is
https://github.com/openvinotoolkit/oneDNN/tree/xiuchuan/dev-v3.6
but without permission of pushing cherry-pick commits to branch
https://github.com/openvinotoolkit/oneDNN/tree/v3.6_for_ie_master

oneDNN fork PR: openvinotoolkit/oneDNN#273

Author: azhai219

src/plugins/intel_cpu/src/nodes/kernels/x64/brgemm_kernel.cpp

@@ -273,6 +273,7 @@ void BrgemmKernel::init_brgemm_copy_a(
                         : DnnlExtensionUtils::sizeOfDataType(static_cast<dnnl::memory::data_type>(dt_in0));
     brgCopyKernelConf.transposed_A = transpose;
     brgCopyKernelConf.isa = is_avx_f16_only ? avx512_core_fp16 : avx512_core_amx;
+    brgCopyKernelConf.orig_wei_dt = static_cast<dnnl_data_type_t>(DnnlExtensionUtils::ElementTypeToDataType(inType));
 
     create_brgemm_matmul_copy_a(brgCopyKernel, &brgCopyKernelConf);
 }

src/plugins/intel_cpu/tests/functional/shared_tests_instances/skip_tests_config.cpp

@@ -479,18 +479,12 @@ std::vector<std::string> disabledTestPatterns() {
         R"(.*smoke_MatMulCompressedWeights_corner_cases_basic/MatmulWeightsDecompression.CompareWithRefs/data_shape=\[\?.\?.\?\]_\(\[1,1,4096\]\)_weights_shape=\[4096,4096\]_group_size=128_weights_precision=nf4_decompression_precision=f16_scale_precision=undefined_transpose_weights=0_decompression_subtract=full_reshape_on_decompression=1_config=\(\).*)");
     retVector.emplace_back(R"(.*smoke_RDFT_CPU_1D/RDFTTestCPU.CompareWithRefs/prec=f32_IS0=\[\]_TS0=\(\(126\)\)_constAxes=true_axes=\(\(0\)\)_isInverse=false.*)");
     retVector.emplace_back(R"(.*smoke_RDFT_CPU_2D/RDFTTestCPU.CompareWithRefs/prec=f32_IS0=\[\]_TS0=\(\(16.38\)\)_constAxes=true_axes=\(\(0.1\)\)_isInverse=false.*)");
-    // Issue: MFDNN-12818
-    retVector.emplace_back(R"(.*smoke_LPT/RecurrentCellTransformation.CompareWithRefImpl/f32_\[1,1,3\]_CPU_f32FQ_X_level=256_.*_FQ_W_level=255.*)");
-    retVector.emplace_back(R"(.*smoke_static/ConvertFqRnnToQuantizedRnn.CompareWithRefs/Type=GRUSequence.*2.5.10.*2.1.4.*2.1.4.*)");
 #endif
     if (!ov::with_cpu_x86_avx512_core()) {
         // on platforms which do not support bfloat16, we are disabling bf16 tests since there are no bf16 primitives,
         // tests are useless on such platforms
         retVector.emplace_back(R"(.*(BF|bf)16.*)");
         retVector.emplace_back(R"(.*bfloat16.*)");
-        // Issue: MFDNN-12818
-        retVector.emplace_back(R"(.*smoke_LPT/RecurrentCellTransformation.CompareWithRefImpl/f32_\[1,1,3\]_CPU_f32FQ_X_level=256_.*_FQ_W_level=255.*)");
-        retVector.emplace_back(R"(.*smoke_static/ConvertFqRnnToQuantizedRnn.CompareWithRefs/Type=GRUSequence.*2.5.10.*2.1.4.*2.1.4.*)");
     }
     if (!ov::with_cpu_x86_avx2()) {
         // MatMul in Snippets uses BRGEMM that is supported only on AVX2 (and newer) platforms