[b-re-w]

Refs #125, #41. User request (via the leader): a logical comparison first, measurement later.

Documentation added: docs/investigations/typedpython-vs-cython-numba-lpython.md.

• Seven axes: type sources and failure behaviour, codegen path and timing, object model, CPython identity, loops and arrays, call boundary cost, and the per-benchmark ceilings for nbody, spectral-norm and fannkuch.
• A conclusion listing where TypedPython wins, where it loses, and the design item that would close each gap, with its identity cost.
• Each claim is tagged [doc], [src], [measured] or [inferred].
• The current state of LPython: alpha, last release v0.22.0 on 2024-07-08, last push 2025-12-11.

Closing items that keep identity:

• depth precheck and fast entry;
• interval proofs for integer checks;
• zero-copy buffer-protocol arrays;
• an exact-type dict/str fast path;
• order-independent parallel loops after free-threading.

Items that need a user decision:

• N-12 GC timing;
• any opt-in fastmath.