Unit tests for L1-to-DRAM resharding don't assert on bad outputs

[esmalTT]

Summary

I discovered that the L1-to-DRAM reshard tests don't actually assert on bad outputs:

tt-metal/tests/tt_eager/python_api_testing/unit_testing/misc/test_reshard.py

Lines 534 to 547 in bd53785

	input_shard_spec = ttnn.ShardSpec(input_shard_grid, input_shard_shape, input_shard_orientation)
	input_mem_config = ttnn.MemoryConfig(input_sharding_scheme, input_buffer_type, input_shard_spec)
	output_shard_spec = ttnn.ShardSpec(output_shard_grid, output_shard_shape, output_shard_orientation)
	output_mem_config = ttnn.MemoryConfig(output_sharding_scheme, output_buffer_type, output_shard_spec)
	input = torch.randn(input_shape).bfloat16()
	input_tensor = ttnn.Tensor(input, ttnn.bfloat16).to(input_layout).to(device, input_mem_config)
	output_tensor = ttnn.reshard(input_tensor, output_mem_config)
	output = ttnn.to_torch(output_tensor)
	passing, output_log = comp_equal(input, output)

Adding the proper assert shows that several test cases are broken.

To close this issue, we should add a proper assert and fix the broken test cases.