Fix greenlet integration with greenlet 3.4.0

[godlygeek]

Our attempts to handle ancient versions of greenlet are causing us to try to import the settrace function from greenlet rather than from greenlet._greenlet, which is failing because we're trying to import it before greenlet's own __init__.py has imported it.

Drop support for greenlet versions older than 1.0, which is more than 5 years out of date at this point, to fix this issue.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.38%. Comparing base (1952221) to head (f739785).
⚠️ Report is 9 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #908      +/-   ##
==========================================
- Coverage   92.49%   92.38%   -0.12%     
==========================================
  Files          99       99              
  Lines       11746    11785      +39     
  Branches      426      429       +3     
==========================================
+ Hits        10865    10888      +23     
- Misses        881      897      +16     

Flag	Coverage Δ
cpp	92.38% <100.00%> (-0.12%)	⬇️
python_and_cython	92.38% <100.00%> (-0.12%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[godlygeek]

Specifically, the full failure mode here is:

• The user starts Memray tracking
• The user does import greenlet
• The greenlet module does from ._greenlet import ...
• The interpreter dlopen's the _greenlet shared library
• Memray's dlopen hook fires and sets a flag to tell our Python profile function to try to install our greenlet trace function the next time a Python instruction is run
• The greenlet._greenlet of greenlet 3.4.0 does a PyImport_ImportModule("atexit") from it's initialization function
• Because atexit is written in Python, our Python profile function runs
• We try to access sys.modules["greenlet._greenlet"] but fail because it's not yet added to sys.modules
• We fall back to using sys.modules["greenlet"] for greenlet 0.x compatibility
• We try to access greenlet.settrace, which fails because it hasn't yet been imported by greenlet from greenlet._greenlet

The PyImport_ImportModule("atexit") was reverted in greenlet 3.5.0, but removing the fallback to import greenlet.settrace when greenlet._greenlet.settrace can't be imported fixes the issue for greenlet 3.4.0 and hardens us against the possibility of the greenlet module running other Python code in the future, or allowing a context switch to another thread (what we were doing was probably broken in free-threading builds as well).

[lkollar]

Would it be worth mentioning that in the news fragment that this drops greenlet support before 1.0? LGTM otherwise.

[godlygeek]

Would it be worth mentioning that in the news fragment that this drops greenlet support before 1.0? LGTM otherwise.

Good point, done.