ensure we set the right value to gc_first_tid (#54645)

This may introduce a correctness issue in the work-stealing termination
loop if we're using interactive threads and GC threads simultaneously.

Indeed, if we forget to add `nthreadsi` to `nthreads`, then we're
checking in the mark-loop termination protocol a range `[gc_first_tid,
gc_first_tid + jl_n_markthreads)` of threads which is "shifted to the
left" compared to what it should be.

This implies that we will not be checking whether the GC threads with
higher TID actually have terminated the mark-loop.

Author: d-netto

src/threading.c

@@ -735,7 +735,7 @@ void jl_init_threading(void)
     jl_atomic_store_release(&jl_all_tls_states, (jl_ptls_t*)calloc(jl_all_tls_states_size, sizeof(jl_ptls_t)));
     jl_atomic_store_release(&jl_n_threads, jl_all_tls_states_size);
     jl_n_gcthreads = ngcthreads;
-    gc_first_tid = nthreads;
+    gc_first_tid = nthreads + nthreadsi;
 }
 
 static uv_barrier_t thread_init_done;

test/gc.jl

@@ -5,11 +5,13 @@ using Test
 function run_gctest(file)
     let cmd = `$(Base.julia_cmd()) --depwarn=error --rr-detach --startup-file=no $file`
         @testset for test_nthreads in (1, 2, 4)
-            @testset for concurrent_sweep in (0, 1)
-                new_env = copy(ENV)
-                new_env["JULIA_NUM_THREADS"] = string(test_nthreads)
-                new_env["JULIA_NUM_GC_THREADS"] = "$(test_nthreads),$(concurrent_sweep)"
-                @test success(run(pipeline(setenv(cmd, new_env), stdout = stdout, stderr = stderr)))
+            @testset for test_nithreads in (0, 1)
+                @testset for concurrent_sweep in (0, 1)
+                    new_env = copy(ENV)
+                    new_env["JULIA_NUM_THREADS"] = "$test_nthreads,$test_nithreads"
+                    new_env["JULIA_NUM_GC_THREADS"] = "$(test_nthreads),$(concurrent_sweep)"
+                    @test success(run(pipeline(setenv(cmd, new_env), stdout = stdout, stderr = stderr)))
+                end
             end
         end
     end