gh-155860: Reject a detached window in panel.replace() - #155861

curses.screen.close() detaches the screen's standard window: the wrapper object stays alive but the curses window behind it is gone. panel.replace() passed that window straight to replace_panel(), so the panel ended up holding a NULL window and curses dereferenced it on the panel's next use.

This adds the missing check before replace_panel(), raising curses.panel.error the way new_panel() already does when handed the same window. new_panel() is left alone: it already rejects the window, so there is nothing to fix there.

The reproducer from the issue now stops at replace() instead of dying:

$ TERM=xterm-256color ./python repro.py; echo "rc=$?"
Traceback (most recent call last):
  File "/home/vyron-vasileiadis/projects/forks/cpython/repro.py", line 7, in <module>
    panel.replace(win)              # accepted, stores a NULL window
    ~~~~~~~~~~~~~^^^^^
curses.panel.error: the window has been detached
rc=1

The new test fails without the change to Modules/_curses_panel.c. With that hunk reverted it reports FAIL, then takes the whole process down in del_panel() at teardown:

test_close_then_panel_replace (test.test_curses.ScreenTests.test_close_then_panel_replace) ... FAIL
  Binary file "/usr/lib/x86_64-linux-gnu/libpanelw.so.6", at del_panel+0x90 [0x734e0319a6d0]

With it, and the module as a whole, plus a refleak run:

test_close_then_panel_replace (test.test_curses.ScreenTests.test_close_then_panel_replace) ... ok

$ ./python -m test test_curses -u curses
Total tests: run=169 skipped=3
Result: SUCCESS

$ ./python -m test test_curses -u curses -R 3:3
Total tests: run=169 skipped=3
Result: SUCCESS

No NEWS entry: screen.close() is new in 3.16 and unreleased, so no released version can reach this crash. The change is news-exempt rather than missing an entry.

• Issue: Segfault when panel.replace() is given a detached window #155860