Use specilized kernel for f-arrays and sum by axis=1. Add keepdims support

[AlexanderKalistratov]

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?
• Have you checked performance impact of proposed changes?
• If this PR is a work in progress, are you filing the PR as a draft?

Tests in #1488 covers this functionality. Except keepdims due to #1487

Speeds up cov with rowvar=False parameter. As it is used in pca workload.

import dpnp
import numpy
import time

a = dpnp.ones((1048576, 128))

dpnp.cov(a, rowvar=False, dtype=a.dtype)
start = time.time()
dpnp.cov(a, rowvar=False, dtype=a.dtype)
print(time.time() - start)

Before:

0.6248629093170166

After:

0.229935884475708

[antonwolfy]

I've checked the PCA workload on the laptop:

	numpy	dpnp CPU	dpnp GPU OCL	dpnp GPU L0	size
baseline	0.61 s	0.75 s	0.71 s	0.68 s	1048576
PR #1489	0.6 s	0.57 s	0.3 s	0.29 s	1048576

Both CPU and GPU are now in targets for the workload, awesome.
I believe it closes #1398.