test_runner: capture process state in entry point instead of runner

The run() function in the test runner was directly accessing process.execArgv,
process.cwd(), and process.env to configure child processes. This is
problematic because run() is part of the public API and should not depend
on ambient process state.

This commit moves those reads to the CLI entry point (test_runner.js) and
passes them as explicit options to run():
- main/test_runner.js: capture process.execArgv, process.env, process.cwd()
- runner.js getRunArgs(): use execArgv parameter instead of process.execArgv
- runner.js run(): default execArgv to process.execArgv for backward compat

Refs: #53867

Author: Krsna Suraj

lib/internal/main/test_runner.js

@@ -31,6 +31,9 @@ if (isUsingInspector() && options.isolation === 'process') {
 }
 
 options.globPatterns = ArrayPrototypeSlice(process.argv, 1);
+options.execArgv = ArrayPrototypeSlice(process.execArgv);
+options.env = process.env;
+options.cwd = process.cwd();
 
 debug('test runner configuration:', options);
 run(options).on('test:summary', (data) => {

lib/internal/test_runner/runner.js

@@ -206,8 +206,9 @@ function getRunArgs(path, { forceExit,
    * An example of such option are --allow-natives-syntax and --expose-gc
    */
   const nodeOptionsSet = new SafeSet(processNodeOptions);
+  const sourceExecArgv = execArgv.length > 0 ? execArgv : process.execArgv;
   const unknownProcessExecArgv = ArrayPrototypeFilter(
-    process.execArgv,
+    sourceExecArgv,
     (arg, i, arr) => !nodeOptionsSet.has(arg) && filterExecArgv(arg, i, arr),
   );
   ArrayPrototypePushApply(runArgs, unknownProcessExecArgv);