Set OPENBLAS_NUM_THREADS=1 on local Distributed workers (#47803)

This should prevent LinearAlgebra from trying to increase our OpenBLAS
thread count in its `__init__()` method when we're not trying to enable
threaded BLAS.

(cherry picked from commit a8b3994)
(cherry picked from commit 8830c26)

Author: staticfloat

stdlib/Distributed/src/cluster.jl

@@ -529,6 +529,7 @@ default_addprocs_params() = Dict{Symbol,Any}(
     :dir      => pwd(),
     :exename  => joinpath(Sys.BINDIR::String, julia_exename()),
     :exeflags => ``,
+    :env      => [],
     :enable_threaded_blas => false,
     :lazy => true)
 

stdlib/Distributed/src/managers.jl

@@ -450,10 +450,18 @@ function launch(manager::LocalManager, params::Dict, launched::Array, c::Conditi
     exename = params[:exename]
     exeflags = params[:exeflags]
     bind_to = manager.restrict ? `127.0.0.1` : `$(LPROC.bind_addr)`
+    env = Dict{String,String}(params[:env])
+
+    # If we haven't explicitly asked for threaded BLAS, prevent OpenBLAS from starting
+    # up with multiple threads, thereby sucking up a bunch of wasted memory on Windows.
+    if !params[:enable_threaded_blas] &&
+       get(env, "OPENBLAS_NUM_THREADS", nothing) === nothing
+        env["OPENBLAS_NUM_THREADS"] = "1"
+    end
 
     for i in 1:manager.np
         cmd = `$(julia_cmd(exename)) $exeflags --bind-to $bind_to --worker`
-        io = open(detach(setenv(cmd, dir=dir)), "r+")
+        io = open(detach(setenv(addenv(cmd, env), dir=dir)), "r+")
         write_cookie(io)
 
         wconfig = WorkerConfig()