Improve consistency in how we pass around dynamic array of elements

[dgomezTT]

vector vs SmallVector vs span: There should be a single way to represent a (dynamic) array of elements. The span is an ideal option as it represents an interface rather than a class, but even std::vector or SmallVector are okay if they are used consistently.

[ayerofieiev-tt]

From @svuckovicTT

ttnn/cpp/ttnn/operations/reduction/generic/generic_reductions.hpp

struct Reduce {
    static Tensor invoke(
        const Tensor& input_tensor_arg,
        const std::optional<std::variant<int, ttnn::SmallVector<int>>>& dim_arg = std::nullopt,    // <---- smallvec here
        const bool keepdim = true,
        const std::optional<MemoryConfig>& memory_config_arg = std::nullopt,
        const std::optional<DeviceComputeKernelConfig>& compute_kernel_config = std::nullopt,
        float scalar = 1.0f);
};

vs

ttnn/cpp/ttnn/operations/data_movement/permute/permute.hpp

struct ExecutePermute {
    static ttnn::Tensor invoke(
        uint8_t queue_id,
        const ttnn::Tensor& input_tensor,
        const SmallVector<int64_t>& dims,  // <---
        const std::optional<MemoryConfig>& memory_config,
        const std::optional<float>& pad_value = 0.0f);

    static ttnn::Tensor invoke(
        const ttnn::Tensor& input_tensor,
        const SmallVector<int64_t>& dims,  // <---
        const std::optional<MemoryConfig>& memory_config,
        const std::optional<float>& pad_value = 0.0f);

    static ttnn::Tensor invoke(
        const ttnn::Tensor& input_tensor,
        const SmallVector<int64_t>& dims,  // <--- 
        const std::optional<float>& pad_value = 0.0f);
};

Basically, what we want to unify to Span instead of SmallVector in op invoke methods.
Please keep in mind, span does not maintain ownership over data. But from the interface standpoint it is convenient to accept span as it allows whatever input, whether its array, vector, smallvector or any other compatible container.

There are cases where a fixed size array is used. Such cases are out of scope of this effort

struct ExecuteUpSample {
    static ttnn::Tensor invoke(
        const ttnn::Tensor& input_tensor,
        std::variant<int, tt::tt_metal::Array2D> scale_factor,  // <---
        const std::string& mode = std::string("nearest"),
        const std::optional<MemoryConfig>& output_mem_config = std::nullopt,
        const std::optional<DeviceComputeKernelConfig>& compute_kernel_config = std::nullopt);
};

CC @dgomezTT