gh-156939: Detect buffer overflow in PyBytesWriter in debug mode

[vstinner]

Reserve one byte in PyBytesWriter used as a canary byte: set it to a special value. PyBytesWriter_Finish() checks if the canary byte has been overriden. Add a test on the feature.

Update buffer overflow check in fcntl: allocate extra guard bytes in the writer and then truncate these bytes.

• Issue: C API: Detect buffer overflow in PyBytesWriter #156939

[vstinner]

Oh, the test_xml_etree_c failure is a legit failure: unicode_encode_ucs1() calls with errors="xmlcharrefreplace" writes a NUL byte after the writer buffer. xmlcharrefreplace() calls sprintf(str, "&#%d;", PyUnicode_READ(kind, data, i)) which writes a final NUL byte.

[vstinner]

The test_struct error is also a legit bug: a Pascal string of 0 characters writes a NUL byte into an empty buffer, and so triggers a buffer overflow.

[vstinner]

On Ubuntu, 8 tests are failing:

8 tests failed:
    test_codeccallbacks test_fcntl test_minidom test_sax test_struct
    test_urllib test_xml_etree test_xml_etree_c

[vstinner]

I fixed struct.pack() in PR gh-157071 and xmlcharrefreplace in PR gh-157109.

[vstinner]

cc @serhiy-storchaka