End to End implementation of ttnn.prod op

[mmanzoorTT]

closes #1790 #1791

[sdjordjevicTT]

It looks good. Please update and elaborate more on the new operation description in the dialect definitions.

[azecevicTT]

I have some concerns regarding restrictions imposed on much higher level of abstraction than the place where they belong to. We shouldn't make a StableHLO depend on TTNN, it's a very live project that changes every day, so even when some of those things get fixed we are still left with an unnecessary restrictions. On top of that we now have an infrastructure set in place to circumvent most (if not all) of those restrictions, so we should use it.

[azecevicTT]

This looks much better, but TTNN modelling is still an issue.

What's your opinion on this, to be quite honest this one

    static Tensor invoke(
        const Tensor& input,
        bool all_dimensions = false,
        int64_t dim = 0,
        const bool keepdim = false,
        const std::optional<MemoryConfig>& memory_config = std::nullopt);

looks terrible to me because of this part

        bool all_dimensions = false,
        int64_t dim = 0

but the second one lacks keepdim (and is also a DPS).

[azecevicTT]

Looks much better now, I left a few comments that should be addressed, but nothing critical.