Preserve the Flow SIGTERM handler in AM callbacks

[robjarawan]

Fixes #163.

The AM gather and send callbacks were replacing the instance SIGTERM handler with SIG_DFL when they were constructed. So a normal SIGTERM killed the process immediately instead of calling Flow.stop_request() and letting the normal on_stop cleanup run.

So what I did was leave the instance handler in place. I added a regression test that starts a real child process, installs sarracenia.instance.instance.stop_signal, constructs each real AM callback, and sends the child an actual SIGTERM.

Before this change, both children reported parent_handler_preserved: false, returned no shutdown result, and exited with -15. After the change, both callbacks preserve the handler, request the Flow stop, close their sockets, and exit with 0.

I also checked this together with the existing partial-send fix from #119. That commit applies cleanly and all 22 combined AM tests pass.

Tests:

• tests/sarracenia/flowcb/am_signal_test.py: 2 passed
• focused AM tests: 20 passed
• combined with fix: send complete AM frames over TCP #119: 22 passed
• broader suite excluding the unavailable Azure fixture: 356 passed, 1 skipped
• local maintenance flow: PASS, 5 messages received and downloaded, queues/configurations cleaned up

This branch also includes the local-fixture CI repair from #140 so the fork maintenance jobs do not depend on public dynamic flows.

Related upstream work: MetPX#1653. I have left that upstream PR unchanged; this draft is based on the current fork development branch and contains the stronger real-signal regression proof.

[github-actions]

Test Results

376 tests   375 ✅  1m 44s ⏱️
  1 suites    1 💤
  1 files      0 ❌

Results for commit 05b1095.