TypedPython: ownership inference, memory and minimal interpreter footprint in the design (D1-D8) #194

What

Record in docs/design/typedpython.md (new §4.3.5) the ownership inference plan, the memory plan against Rust and the minimal-interpreter plan, with the maintainer's decisions of 2026-10-05:

• D1 yes: identity (is, id) of exact int/float values read from slots is outside the equivalence guarantee (the language leaves identity of immutable values to the implementation).
• D2 yes, conditional: for @compiled classes the compiler may create the type (PyType_FromSpec); differential tests prove __module__, __qualname__, __slots__, MRO, pickle and inspect match the class statement.
• D3 yes, conditional: stores into slots of @compiled classes enforce the annotated type (TypeError at the store).
• D4 no: no non-GC classes for now (saves 0 B on 3.15t), so no final restriction.
• D5: defect, N-12 borrowed field reads are unsound on free-threaded 3.15t #193.
• D6 yes for release app builds only: site off; pypackpack sets sys.path; dev mode keeps it.
• D7 yes: patch our own FT CPython builds to return mimalloc memory on platform memory pressure, minimal patch, proposed upstream.
• D8 no for now: no C-init-code for class/function definitions.
• No decision needed: PEP 810 lazy import in pythonx and generated code; memory (physical footprint) added to the benchmarks.

The investigation goes to docs/investigations/typedpython-ownership-memory.md.

Done when

§4.3.5 added, §4.3.4 FT statements corrected per #193, §7 and §8 updated, investigation recorded, no em-dash. Docs only.