gh-103295: fix stack overwrite on 32-bit in perf map test harness

[carljm]

We can't use K format specifier in PyArg_ParseTuple to write to a void *; it always writes 64 bits, so writes too much on 32-bit systems. Use O instead and an explicit call to PyLong_AsVoidPtr to get the pointer value.

Also fix the error check so it detects either error code (-1 or -2) and raises a more informative error using errno. I verified this error handling by temporarily changing the perf map file path to a nonexistent directory, and the test raised FileNotFoundError as expected.

Also simplify the return value to just use PyLong_FromLong; we don't need to go through Py_BuildValue for such a simple case.

• Issue: expose API for writing perf map files #103295

A Fedora build with this patch applied on top of 3.12.0b1 tarball.

https://koji.fedoraproject.org/koji/taskinfo?taskID=101500963

It is hopefully set up not to terminate the build immediately when one architecture fails. Let's see.

Looks like the builds all succeeded 🎉

A Fedora build with this patch applied on top of 3.12.0b1 tarball.

https://koji.fedoraproject.org/koji/taskinfo?taskID=101500963

All architectures built, tests have passed.

I'm going to go ahead and merge this to help unblock the Fedora builds, since this PR only touches test harness and the test signal is all good (including the Fedora build signal on multiple platforms, including 32-bit.)

If there are any concerns with anything here, I'll be happy to do a follow-up PR to address comments.

Thanks @carljm for the PR 🌮🎉.. I'm working now to backport this PR to: 3.12.
🐍🍒⛏🤖

GH-104823 is a backport of this pull request to the 3.12 branch.