#0: fix bug in createqkv

[kpaigwar]

Desciption

This fixes the failure for llama3/test/test_llama_attention.py in CI introduced by following PR for slicing support in ttnn.nlp_create_QKV_heads_decode when one of the optional tensor argument buffer is read even though its value doesn't exist.

Error

https://github.com/tenstorrent/tt-metal/actions/runs/13076612535/job/36490477945

ttnn::tensor::buffer(): Expected Tensor with DeviceStorage, got StorageType::OWNED
backtrace:
 --- void tt::assert::tt_throw_impl<fmt::v11::basic_format_string<char, tt::tt_metal::StorageType const&>, tt::tt_metal::StorageType>(char const*, int, char const*, char const*, fmt::v11::basic_format_string<char, tt::tt_metal::StorageType const&> const&, tt::tt_metal::StorageType const&)
 --- /home/ubuntu/actions-runner/_work/tt-metal/tt-metal/ttnn/ttnn/_ttnn.so(+0x5123de) [0x7f21d2be33de]
 --- ttnn::operations::experimental::transformer::multi_core_nlp_create_qkv_heads_decode_sharded_input(tt::tt_metal::Tensor const&, unsigned int, unsigned int, unsigned int, bool, std::__1::optional<tt::tt_metal::Tensor const> const&, std::__1::optional<unsigned int const>, std::__1::vector<tt::tt_metal::Tensor, std::__1::allocator<tt::tt_metal::Tensor>>&, tt::umd::xy_pair)