Delete convd_host_weights and update all tests using conv2d

[LPanosTT]

We have ops prepare_conv_weights_for_ttnn and prepare_conv_bias_for_ttnn that will pre-format convolution weights and biases for conv2d (without pushing them to device). In the previous implementation, you would pass the NHWC weights to ttnn.conv2d and ttnn.conv2d would return the formatted weights/bias on device for you to use in subsequent calls. The two aforementioned ops do this beforehand now.

TASKS

• ttnn.conv2d is now pybinded to conv2d_host_weights which is the previous conv2d implementation. This needs use conv2d now
• All tests that use ttnn.conv2d must use ttnn.prepare_conv_weights_for_ttnn and ttnn.prepare_conv_bias_for_ttnn to format weights/biases beforehand.

[shwetankTT]

Waiting for the merge of this PR. https://github.com/tenstorrent/tt-metal/pull/14566