Correlation via fft implementation

[AlexanderKalistratov]

• Added keyword method to correlate function similar to scipy correlate
• If method == 'auto' method is choosing automatically between direct and fft
• Added implementation of fft-based correlation
• fft-based implementation may have accuracy issues, so it is validated in non-standard way.

Depends on: #2180, #2202

[github-actions]

View rendered docs @ https://intelpython.github.io/dpnp/pull/2203/index.html

[oleksandr-pavlyk]

Perhaps it is better to do:

Suggested change

	assert a.size >= v.size
	if a.size < v.size:
	raise ValueError("a.size must not be smaller than v.size")

[oleksandr-pavlyk]

These assertions should be turned to checks that raise exceptions if unmet.

This is because Python interpreter allows to turn off assertion checks with -O option:

(dev_dpctl) opavlyk@opavlyk-mobl:~/repos/dpctl$ python -O -c "assert False"
(dev_dpctl) opavlyk@opavlyk-mobl:~/repos/dpctl$ python  -c "assert False"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
AssertionError

[AlexanderKalistratov]

These are an internal functions which are supposed to be called from already validated context.
These asserts are more reminders than an actual checks

I could convert them into checks but this would also decrease coverage statistics

[oleksandr-pavlyk]

Since math.ceil returns float, it should be converted to integer:

Suggested change

	fft_size = 2 ** math.ceil(math.log2(padded_size))
	fft_size = 2 ** int(math.ceil(math.log2(padded_size)))

[coveralls]

coverage: 65.17% (+0.09%) from 65.08%
when pulling f7df1ac on correlate_fft
into 4d1a6a9 on master.