rust(bug): fix python bindings segfault

[solidiquis]

rust(fix): stop the sift-stream-bindings tokio runtime before python exit

Relevant issues:

• Do we need some kind of shutdown method? PyO3/pyo3-async-runtimes#40
• [BUG] future_into_py can call a closed asyncio loop during cancellation PyO3/pyo3-async-runtimes#90

Problem

User reported segfault. Happens consistently in the tokio-rt-worker thread. Happens when futures finish
on the tokio side but Python interpreter already shut down.

Fix

The bindings now own their tokio runtime (src/runtime.rs) and plug it into pyo3_async_runtimes::generic, so task locals, cancellation, and future resolution are unchanged.
All sixteen future_into_py call sites route through it.

• New shutdown(timeout=5.0) stops the runtime and joins its threads with the GIL released. The module registers it with atexit at import, so a normal interpreter exit always runs it while the interpreter is still intact. It can also be called directly, for example before os._exit().
• New is_shut_down() reports whether it has run. Calls that need the runtime afterwards raise RuntimeError instead of crashing.
• Runtime threads are named sift-stream-worker so future crash logs identify them.

Verification

User provided a repro script which was used for verification.

Linux container (Debian 13, aarch64, CPython 3.11.16, glibc 2.41), running the user's repro script unchanged:

Build	User repro, 150 runs	Deterministic repro, 20 runs
main (0.5.1)	134 clean, 10 SIGABRT, 6 SIGSEGV	20 clean, pyo3 panics on every worker
this branch	150 clean	20 clean, no stderr

• Bindings test suite: 63 passed on Linux 3.11, macOS 3.11, and macOS 3.13.
• cargo clippy --all-targets -D warnings clean. Generated .pyi stub unchanged.

Releases

• sift-stream-bindings 0.5.1 to 0.5.2, changelog entry under Unreleased in rust/CHANGELOG.md.
• Once this is merged we will release a new version of the bindings then follow-up with another PR bumping the Python version and release that.