Add support for binary SFPU OPs

[ttmtrajkovic]

Need to add support for binary SFPU OPs that will accept 2 input tiles and produce one output.
both input tiles need to reside in DEST, thus effective size of subblock will be reduced by 2 for these ops.

@rdjogoTT to add more details as the work progresses
Currently gating other LLK and OP issues:

• Pow operation needs support for Tensor exponent variant #13857
• Get add_tile API support for int32 values  #13754
• Binary Bitwise ops #13582
• Full FP32 precision eltwise add, for reduction of gradients across model replicas #13445

Milos

[rdjogoTT]

Update can be found at: #13857 (comment)
Basically, LLK level Binary SFPU OP support has been added for WH with APIs for basic OPs working (Add, Sub, Mul), and Pow is still in progress.

[rdjogoTT]

Update here: #13857 (comment)
-> Pow PCC >0.99998 for non-negative base values, working on handling special cases where base is negative or base/power are NaN or Inf.