Support floating-point exponents for ttnn.pow

[jdh8]

As we found in tenstorrent/pytorch2.0_ttnn#211, ttnn.pow only supports integer exponents. Even if we only need integral support (as current statistics say), we can still optimize current implementation with binary exponentiation.

tt-metal/tt_metal/hw/ckernels/blackhole/metal/llk_api/llk_sfpu/ckernel_sfpu_power_iterative.h

Lines 18 to 26 in 021137c

	for (int d = 0; d < 8; d++) {
	vFloat in = dst_reg[0];
	vFloat result = 1.0f;
	for (uint i = 0; i < exponent; i++) {
	result *= in;
	}
	dst_reg[0] = result;
	dst_reg++;
	}

[mouliraj-mcw]

Hi @jdh8

• The pow function with floating-point exponents is already implemented in the composite structure
  I would like some clarification on the requirement here - is it to optimize the current ckernel for pow using binary exponentiation ?

cc: @eyonland