Inline values array into the object

[markshannon]

Feature or enhancement

Currently when allocating a plain Python object, we allocate the object, and its values array.
This has a few downsides:

1. We perform two allocations instead of one.
2. An extra indirection is needed when accessing an attribute on the object
3. If the __dict__ is materialized we can no longer use specialized lookup.

We could fix three with an extra pointer in the object header, but that would waste more space.

We should append the values array directly after the object header, which fixes the above issues.

It adds some complexity, as we need to track ownership of the values so that they are freed exactly once, but may enable some simplifications as well. Overall, it would seem to make little difference to complexity.

See faster-cpython/ideas#72 for more discussion.

Linked PRs

• GH-115776: Embed the values array into the object, for "normal" Python objects. #116115
• GH-115776: Static object are immortal, so mark them as such. #117673
• GH-115776: Allow any fixed sized object to have inline values #123192
• GH-115776: Rename Py_TPFLAGS_INLINE_VALUES as _Py_TPFLAGS_INLINE_VALUES #128635

[encukou]

This change introduced a regression, see #117750

[encukou]

The new public API, Py_TPFLAGS_INLINE_VALUES, should be documented.

[cdce8p]

Posting it here in case it was missed. #123192 caused mypyc to break. We haven't been able to narrow it down to a clean reproducer just yet unfortunately. For more details see python/mypy#17973