[FEA] Relax pe_count for CSP vCPU with hard limits

[pdmack]

Is your feature request related to a problem? Please describe.
Azure subscriptions impose a hard limit on vCPU resources in their VM.
So if we run a Morpheus pipeline with 8 threads on a NC6s v2 (6 vCPU) instance we get an abort in thread_engines.cpp

F20220809 15:38:51.809185   262 thread_engines.cpp:38] 
Check failed: launch_options().pe_count == m_cpu_set.weight() 
(8 vs. 6) mismatch in the number of cores in the cpu set with respect to the requested pe_count
*** Check failure stack trace: ***
Aborted (core dumped)

Describe the solution(s) you'd like

• Graceful degradation to the value of m_cpu_set.weight() with appropriate warnings in logging.
• Possibly make hard v. soft limits configurable via the SRF API and throw the error back up to Morpheus (app layer).

Describe alternatives you've considered
Workaround is for the Morpheus pipeline to be run with threads <= vCPU count in Azure VM.

Additional context
https://docs.microsoft.com/en-us/azure/databricks/kb/clusters/azure-core-limit

[pdmack]

Actually this is not solely an Azure issue. I have reproduced this in EC2 and GCP and bare metal (80 cpu).
Basically anywhere Morpheus asks for more threads than CPU, vCPU or otherwise.

[jarmak-nv]

• @mdemoret-nv to provide a prototype API from Morpheus to understand thread availability
• @ryanolson to expose the api up through SRF