[CPU] Conv-DWConv-PRelu fusing fix (#33917)
### Details:
- *Kernel `jit_avx2_1x1_conv_kernel_f32_old` skips DW Convolution
post-op on initialization step, but tries to add it on the code
generation step. This fix aligns behavior.*
- *ConvDWConv test does not work as expected: it does not check any
fusing. There is a condition in the `FuseConvolutionAndDWConvolution`:
```(dw_conv_input_size + dw_conv_output_size > L3_cache_size / 2)```
that is not satisfied in test, but test is green. Fix is adding fusing
check via `CpuTestWithFusing` class.*
- Fusing `FuseConvolutionAndDWConvolution` is applicable for AVX2 only,
thus this code was disabled for non-X86 platforms to reduce binary size.
Related OneDNN PR: https://github.com/openvinotoolkit/oneDNN/pull/296
### Tickets:
- *173761*