Improve handling of floating-point parameters

[psyhtest]

Floating-point parameters (like a scalar scaling factor) were more or less an afterthought in CK-NNTest. Therefore, this area is still rather buggy.

For example, with two GEMM parameters alpha=1.0 and beta=0.0, the following fails:

$ ck run program:gemm-pytorch --dataset_file=shape-1024-1024-1024-1.0-0
...
CK error: [program] problem opening json file=/homes/anton/CK/ck-nntest/dataset/tensor-gemm-0001/shape-1024-1024-1024-1.json ([Errno 2] No such file or directory: '/homes/anton/CK/ck-nntest/dataset/tensor-gemm-0001/shape-1024-1024-1024-1.json')!

while two equivalent representations work fine:

$ ck run program:gemm-pytorch --dataset_file=shape-1024-1024-1024-1.0-0.0
...
  (checking output correctness ...)
     * Searching directory with reference output "default-f4407088898d5f10-1.0-0.0-1024-1024-1024-1-42" ...
     * Reference output found - validating ...
     * File: /disk1/homes/anton/CK/local/program.output/program-uid-6ac244466493cbcd/default-f4407088898d5f10-1.0-0.0-1024-1024-1024-1-42
       Validated successfully!

$ ck run program:gemm-pytorch --dataset_file=shape-1024-1024-1024-1-0
...
  (checking output correctness ...)
     * Searching directory with reference output "default-f4407088898d5f10-1.0-0.0-1024-1024-1024-1-42" ...
     * Reference output found - validating ...
     * File: /disk1/homes/anton/CK/local/program.output/program-uid-6ac244466493cbcd/default-f4407088898d5f10-1.0-0.0-1024-1024-1024-1-42
       Validated successfully!

The dataset files are completely identical:

$ cd `ck find dataset:tensor-gemm-0001`
$ md5sum shape-1024-1024-1024*.json
90a74c27b382bbd97e390a8e1d37de3b  shape-1024-1024-1024-1.0-0.0.json
90a74c27b382bbd97e390a8e1d37de3b  shape-1024-1024-1024-1.0-0.json
90a74c27b382bbd97e390a8e1d37de3b  shape-1024-1024-1024-1-0.json
$ cat shape-1024-1024-1024-1-0.json 
{
  "CK_GEMM_ALPHA": 1.0, 
  "CK_GEMM_BETA": 0.0, 
  "CK_GEMM_K": 1024, 
  "CK_GEMM_M": 1024, 
  "CK_GEMM_N": 1024
}