Change number of streams in latency mode on XEON to 1

[wangleis]

Details:

• Change number of streams in latency mode on XEON to 1
• Update CPU unit test

Tickets:

• CVS-129895

[dmitry-gorokhov]

@wangleis I tried this PR and see the following behavior:

1. If I specify hint latency only infer happens on single socket.
2. If I additionally to latency hint specify nstreams=1 it starts to use both sockets.
   I would say this is not expected behavior. We should use single socket by default for any case. If user wants to use both socket, then nthreads should be manually specified.
   Alternativelly we might consider exposure of LatencyThreadingMode as cpu property to make set up process more convenient for workloads where whole systems should be occupied.

[wangleis]

@wangleis I tried this PR and see the following behavior:

1. If I specify hint latency only infer happens on single socket.
2. If I additionally to latency hint specify nstreams=1 it starts to use both sockets.
   I would say this is not expected behavior. We should use single socket by default for any case. If user wants to use both socket, then nthreads should be manually specified.
   Alternativelly we might consider exposure of LatencyThreadingMode as cpu property to make set up process more convenient for workloads where whole systems should be occupied.

@dmitry-gorokhov Updated. Please try again.