Use a private duplicate of stdout/stderr fd in CliRunner

[charlieleith]

fixes #3384

Summary

Click 8.3.3 (#3244) started exposing the original stream's file descriptor via _NamedTextIOWrapper.fileno(), so that C-level consumers like faulthandler and subprocess could obtain a real fd from sys.stdout / sys.stderr inside CliRunner isolation.

That regressed any command that performs os.dup2 against the value returned by sys.stdout.fileno() — logging tees, build tools, subprocess wrappers, etc. The redirect happens on the host process's real stdout fd, which clobbers any outer fd-based capture (most notably pytest's capfd/capsys capture pipe). The reporter's repro fails during pytest's own teardown with OSError: [Errno 29] Illegal seek because pytest tries to seek(0) on the tmpfile that has just been overwritten with a non-seekable pipe.

This mirrors how pytest itself handles the same problem in capfd: os.dup the original fd inside CliRunner and expose the duplicate via fileno(). os.dup2 against the duplicate redirects only this private copy, so the host's stdout/stderr — and any capture installed on top of it — stays intact. The duplicates are closed when the isolation context exits.

The semantics from #3244 are preserved: faulthandler.enable(), subprocess, signal handlers, and similar C-level consumers still get a valid, terminal-bound fd.

Repro

The repro from the issue passes after this change:

import io, os, sys
import click
from click.testing import CliRunner

@click.command()
def cmd():
    stdout_fd = sys.stdout.fileno()
    r, w = os.pipe()
    os.dup2(w, stdout_fd)
    os.close(w); os.close(r)
    click.echo("hello")

def test_invoke():
    runner = CliRunner()
    result = runner.invoke(cmd)
    assert result.exit_code == 0

Tests

• Adds test_dup2_over_stdout_fd_does_not_leak in tests/test_testing.py. Fails on main with OSError: Illegal seek during pytest teardown (same backtrace as the issue), passes with this change.
• Existing test_faulthandler_enable still passes — duped fd remains valid for faulthandler.
• Full suite: 1437 passed, 23 skipped, 1 xfailed.

Checklist

• Add tests that demonstrate the correct behavior of the change. Tests fail without the change.
• Add or update relevant docs, in the docs folder and in code (_NamedTextIOWrapper docstring + versionchanged).
• Add an entry in CHANGES.rst summarizing the change and linking to the issue.
• Add versionchanged entries in any relevant code docs.