[nanvix] E: Phase 2 — build 15 Tier-2 stdlib extensions as .so

[esaurez]

Summary

Phase 2 of the .a → .so migration (see nanvix-todo/cpython-static-to-shared-migration.md section 6). Promotes 15 modules with bundled-in-cpython C dependencies from statically linked into python.elf to dlopen-loaded shared objects under lib/python3.12/lib-dynload/.

Modules moved to *shared*

Group	Modules
Async + time (2)	_asyncio, _datetime
Decimal + XML (3)	_decimal, pyexpat, _elementtree
Hashing (5)	_md5, _sha1, _sha2, _sha3, _blake2
Net + IO + posix (5)	select, _socket, _posixsubprocess, fcntl, termios

Bundled-lib trade-off in this PR

Four of these modules carry bundled C library dependencies:

• _decimal → libmpdec.a
• pyexpat + _elementtree → libexpat.a
• _sha2 → libHacl_Hash_SHA2.a

This PR ships them with the bundled archive embedded in the .so itself via the existing CPython per-module LDFLAGS mechanism. The result is functionally complete but the .so files are larger than ideal: _decimal.so is 1593 KB (libmpdec dominates), _sha2.so is 1766 KB (libHacl_Hash_SHA2 dominates), pyexpat.so is 731 KB (libexpat dominates).

Optimizing these four to resolve bundled-lib symbols at dlopen time against python.elf's .dynsym (analogous to the Phase 1B → Phase 1B-drop-libm follow-up) is deferred to a separate PR. The optimization requires either a two-pass configure (build the bundled libs first so they exist when conftest runs) or a linker-script-based approach, neither of which is trivial. Importantly:

• The visibility-merge mechanism that required esaurez/nanvix#26 for libm does NOT apply here — libmpdec / libexpat / libHacl_Hash_SHA2 are pure C with no Rust compiler_builtins shadows. The deferred follow-up is a pure size optimization, not a correctness fix.
• mpdec / expat / HACL have no per-process state concerns (mpdec is pure compute, expat is parser state local to each parser instance, HACL is hashing). The duplicated copies don't have correctness issues like a duplicated libm would.

For the other 11 modules with no bundled deps (_asyncio, _datetime, _md5, _sha1, _sha3, _blake2, select, _socket, _posixsubprocess, fcntl, termios), .so size is what it should be — CPython glue only.

Size impact

Metric	Before	After	Delta
python.elf	17.48 MB	16.67 MB	−0.81 MB
Total Phase 2 .so	0	~6.4 MB	+6.4 MB (much of this is the bundled-lib duplication to be optimized later)

Test coverage

New phase2_snippet in .nanvix/test.py imports each module, asserts non-builtin status, exercises one trivial API call to confirm dlopen + PyInit_<name> succeeded (_asyncio.Future, _decimal.Decimal('1.1') + Decimal('2.2') == Decimal('3.3'), pyexpat.ParserCreate, hash construction, socket creation, etc.), and prints the resolved __file__ path. Phase 1A/1B/1C probes retained.

Validation

Tested on phase0-llfix toolchain overlay:

• All 15 .so files installed under lib-dynload/.
• nm python.elf no longer shows PyInit_<name> for any of the 15.
• python.elf size: 17.48 MB (Phase 1C) → 16.67 MB (Phase 2), −0.81 MB.
• Hello + Phase 1A + 1B + 1C + 2 import probes + lxml + HTTP smoke + full regrtest 160/160 PASS in standalone mode.
• _decimal.Decimal('1.1') + Decimal('2.2') == Decimal('3.3') works (validates libmpdec embed).
• pyexpat.ParserCreate() works (validates libexpat embed).
• _sha2.sha256() works (validates libHacl_Hash_SHA2 embed).

Prerequisites

Stacked on Phase 1C (esaurez/cpython#8). No additional nanvix / newlib / gcc PRs needed beyond what Phase 1 already requires.

Risk

Mechanical configuration change — 15 entries added to *shared* block of Setup.local, same pattern as Phase 1A/1B/1C. The bundled-lib .so size cost is documented and a follow-up optimization is identified.