implement dpnp.mean

[vtavana]

In this PR, dpnp.mean is implemented using its dpctl counterpart.

• 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?

[github-actions]

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

[antonwolfy]

We need also to secure performance against the previous implementation of dpnp.mean on PVC and Iris Xe. Could you please check that here is no performance degradation comes with the new implementation?

[vtavana]

Performance on PVC (Intel(R) Data Center GPU Max 1550):
Timing is obtained using %timeit -r 7 -n 100 for three different cases:

1. a = dpnp.arange(1024*32).reshape(1024,32)

Function	Old implementation	New Implementation
mean(a)	1.98 ms ± 128 µs	1.28 ms ± 44.4 µs
mean(a, axis=0)	2.39 ms ± 26.6 µs	1.52 ms ± 907 µs
mean(a, axis=1)	1.97 ms ± 269 µs	1.37 ms ± 494 µs

2. a = dpnp.arange(1024*32).reshape(32,1024)

Function	Old implementation	New Implementation
mean(a)	1.89 ms ± 150 µs	1.3 ms ± 33 µs
mean(a, axis=0)	2.41 ms ± 48.3 µs	1.17 ms ± 34.1 µs
mean(a, axis=1)	2.33 ms ± 135 µs	1.18 ms ± 26.9 µs

3. a = dpnp.arange(300*300).reshape(300,300)

Function	Old implementation	New Implementation
mean(a)	2.21 ms ± 78.3 µs	1.29 ms ± 58.5 µs
mean(a, axis=0)	2.4 ms ± 28.9 µs	1.15 ms ± 29.8 µs
mean(a, axis=1)	2.39 ms ± 30 µs	1.16 ms ± 34 µs

[vtavana]

Performance on Iris Xe (Intel(R) Graphics [0x46a8]):
Timing is obtained using %timeit -r 7 -n 100 for three different cases:

1. a = dpnp.arange(1024*32).reshape(1024,32)

Function	Old implementation	New Implementation
mean(a)	2.13 ms ± 265 µs	621 µs ± 93 µs
mean(a, axis=0)	1.96 ms ± 127 µs	673 µs ± 102 µs
mean(a, axis=1)	1.97 ms ± 122 µs	674 µs ± 71.4 µs

2. a = dpnp.arange(1024*32).reshape(32,1024)

Function	Old implementation	New Implementation
mean(a)	1.9 ms ± 101 µs	574 µs ± 87 µs
mean(a, axis=0)	1.99 ms ± 55.3 µs	737 µs ± 98.2 µs
mean(a, axis=1)	1.97 ms ± 105 µs	656 µs ± 85.1 µs

3. a = dpnp.arange(300*300).reshape(300,300)

Function	Old implementation	New Implementation
mean(a)	1.92 ms ± 147 µs	622 µs ± 74.5 µs
mean(a, axis=0)	1.98 ms ± 113 µs	654 µs ± 103 µs
mean(a, axis=1)	1.96 ms ± 108 µs	698 µs ± 122 µs