Add an Unstable C API tier as per PEP 689.

• Update CPython docs
• Update the Devguide (PEP 689: Add docs for unstable C API devguide#1060)
• Add initial unstable API from the PEP

Other candidates for inclusion are below.
These usually need discussion first. This is a checklist for having the discussion.
Please try to not hold the individual discussions in this issue.

• Dict watching API
• Functions from PEP-523 not specified in PEP-689 (_PyEval_EvalFrameDefault & co.) -- this is a can of worms tho
• FrameStack API (Add "unstable" frame stack api #91371)
• _Py_HashDouble, see PEP 689 -- Add an unstable C-API tier #91744 (comment)
• PyBytesObject.ob_shash sounds like a good candidate: see https://discuss.python.org/t/15108 (guess fields need a slightly different naming convention though. Sigh. That should have been in the PEP.)
• non-opaque access to frame structs and any other key APIs needed to implement alternate eval loops with comparable performance to the default eval loop (unless & until we can figure out stable public APIs that can deliver equivalent performance) - see Nick's reply
• C APIs that provide access to compiled code whether in AST or opcode form (the API itself may be stable, but the compiled code isn't) - see Nick's reply
• PyLong_FromByteArray/PyUnstable_LongToBase30Digits: https://discuss.python.org/t/20045 [there's now `PyLong_{From,To}NativeBytes
• Anything that starts with an underscore and is documented
• Fast access to PyLong contents [PyLong_Export]

Got any more?
I offer to move other API to the unstable tier myself, if there's a good usage example to base docs & regression tests on [edit 2026:] and a C API WG decision. Please discuss new ideas on Discourse.

Linked PRs

• gh-101101: Unstable C API tier (PEP 689) #101102
• gh-101101: Fix test_code_extra to reset value for refleak test #102350