[ hgemm ] Support beta case @open sesame 06/04 13:13

This commit allows hgemm to get beta condition as well.
Note that beta for here is as follows:
```
C = alpha * A * B + beta * C
```

In many cases, beta is 0.F. However, we can fuse calculation process if we have to do something like:


Tensor A, B, C;
Tensor tmp;
bool transA = false;
bool transB = false;
float scalar = 1.1F // anything you want

// redundant case 1 tmp = A.dot(B); C.multiply_i(scalar); C = tmp.add(C);

// use beta C = A.dot(B, C, transA , transB, scalar);

// redundant case 2 C = A.dot(B); C.add_i(scalar);

// use beta C = C.ones(); C = A.dot(B, C, transA , transB, scalar);


- In addition add zero-init code for beta = 0.F case. According to recent model profiling result, even for initialization, minimizing instruction is quite helpful for overall model latency reduction.

**Self evaluation:**
1. Build test:     [X]Passed [ ]Failed [ ]Skipped
2. Run test:     [X]Passed [ ]Failed [ ]Skipped

Signed-off-by: skykongkong8 <ss.kong@samsung.com>

nnstreamer / nntrainer

[ hgemm ] Support beta case @open sesame 06/04 13:13 #2585