Implement [-t|--threads] command line argument for threads

[fredrikekre]

Implement [-t|--threads] command line argument for specifying the number of Julia threads on startup, fixes #26889.

Regarding #26889 (comment): Currently this does not propagate to workers processes, but this commit enables that: c11d9db

Personally I feel like either case is fine -- for fine grained control you need to use addprocs anyway.

Edit: I updated the PR to propagate number of threads to processes spawned with --procs and/or --machine-file. If you use those arguments in combination (e.g. julia -p2 -t2) is is likely because you want it to propagate. And as before, if you need some other behavior you can customize all you want with addprocs.

[fredrikekre]

Specific questions for triage:

• How should $ julia --procs 2 --threads 3 behave? Currently 1+ 2 processes with 3 threads each.
• How should $ JULIA_NUM_THREADS=4 julia --procs 2 --threads 3 behave? Currently 1+ 2 processes with 3 threads each. I think it makes sense that the command line flag takes precedence since it is more "local".
• What should happen when JULIA_NUM_THREADS are set on the remote machine?

[StefanKarpinski]

How about --threads always takes precedence; if someone wants to start a different number of threads on each machine, they can set JULIA_NUM_THREADS on each machine without passing --threads to each process. We should also add a feature to the machine file format so that you can indicate how many processes and threads per process on each machine.

[JeffBezanson]

Triage is ok with what's implemented here. We also might want to eventually phase out the environment variable.

[oschulz]

Closes #18074