ConvolutionInputGenerator is broken if OFMDim > 1

[ThibautMarty]

Hello,

I found out that the ConvolutionInputGenerator function does not work correctly if OFMDim > 1.
It is used in ConvLayer_Batch and makes the convolution test fails (using test_conv3.tcl).

In ConvLayer_Batch, the stream convInp generated by ConvolutionInputGenerator seems erroneous (spotted by forcing all pixels to zeroes and looking at streams: this one is not full of zeroes).
In ConvolutionInputGenerator, I think that the current_line_in_block index makes an out-of-bound access to the array inputBuf in slidingwindow.h.

Steps to reproduce

On dev branch (e2932ad):

1. replace ofm_dimension = 1 by ofm_dimension = 2 in tb/data/gen_weigths.py
2. generate the weights and config with: pushd tb/data/; python gen_weigths.py; popd
3. run export FINN_HLS_ROOT=$(realpath .)
4. run the test with: pushd tb; vitis_hls -f test_conv3.tcl

The test used test_swg.tclstill works.

I used Vitis HLS 2023.2, but I was working on FINN-HLSLIB using a regular compiler for my project and the test fails too.