gh-158841: Fix crash in pickle.load() when read() returns a bytes subclass

[drakeo338]

pickle.load() crashes when a file object's read() returns a bytes subclass and the data is large enough to need several reads. _Unpickler_ReadFromFile() grows the result with _PyBytes_Resize(), which is only valid for exact bytes. The fix copies a subclass instance into an exact bytes object first. A regression test and a NEWS entry are included.

The new test segfaults on a debug build without the fix and passes with it. test_pickle, test_pickletools, test_copyreg and test_picklebuffer pass.

AI disclosure: written with Claude Code assistance; I reviewed the change and ran the tests.

Fixes #158841.

• Issue: _pickle crashes when read() returns a bytes subclass for large payloads #158841

AI-assisted: I used Claude Code to help write this change. I reviewed it and ran test_pickle, test_pickletools, test_copyreg and test_picklebuffer on a debug build.

[error-3317]

Could we verify that every possible path to _PyBytes_Resize() has an exact bytes object at that point? The new conversion is guarded by cursize < n, which is probably sufficient for the current chunked-read path, but the safety invariant is that _PyBytes_Resize() must never receive a bytes subclass.

[drakeo338]

Four _PyBytes_Resize() calls exist in _pickle.c, and three act on bytes the pickler or load_counted_binbytes allocates itself. The fourth, in _Unpickler_ReadFromFile, runs only inside while (cursize < n), and the new conversion uses the same condition, so a subclass becomes exact bytes before any resize. A non-bytes read() result, such as a bytearray, already raised SystemError and is unchanged.

[drakeo338]

Closing, since the decision is to keep the documented behaviour. Thanks for the review.