gh-154997: Guard stale self->raw access in bufferedio.c after reentrant detach()

[BHUVANSH855]

Summary

Fix stale self->raw accesses in buffered I/O objects after a reentrant
detach().

Introduce a shared raw_access_safe() helper and use it before dispatching
to the underlying raw stream from the remaining affected buffered I/O paths.
This preserves the existing detached and uninitialized error semantics while
avoiding dereferences of a stale self->raw pointer.

The guarded call sites include:

• _io__Buffered_close_impl()
• _buffered_raw_tell()
• _buffered_raw_seek()
• _io__Buffered_truncate_impl()
• _bufferedreader_raw_read()
• _bufferedreader_read_all()
• _bufferedwriter_raw_write()

Tests

Added regression tests covering reentrant detach() during:

• close()
• raw read
• read_all()
• raw write
• truncate()
• post-truncate raw_tell()

These tests verify that the affected operations raise the expected exception
(or preserve the intended behavior for raw_tell()) instead of accessing a
stale raw stream.

The reproducer from the issue was also used to verify the patched code paths.

Notes

raw_seek() was intentionally left unchanged from a testing perspective.
While it now benefits from the shared guard, it was not one of the original
SIGSEGV cases and is outside the scope of this change.

During testing, test_read1_error_does_not_cause_reentrant_failure was found
to abort under ASan even on an unmodified main checkout due to unrelated
allocator behavior, so it is not treated as a regression introduced by this
patch.

• Issue: NULL pointer dereference in BufferedIO methods after re-entrant detach() #154997