Widen pexpect close/terminate delay to avoid free-threaded teardown race

[blink1073]

References

Surfaced by oct2py's CI on the test (3.14t) matrix leg:
https://github.com/blink1073/oct2py/actions/runs/30259504522/job/89956026163

Description

On the free-threaded (no-GIL) Python 3.14t build, OctaveEngine consumers
occasionally see pexpect.exceptions.ExceptionPexpect: Could not terminate the child raised out of REPLWrapper.terminate() during teardown, even
though the Octave subprocess was in fact terminating normally.

Root cause: ptyprocess.PtyProcess.close() closes the fd, sleeps
delayafterclose (default 0.1s), then checks isalive() before raising.
Under the free-threaded build, process reaping/signal delivery appears to
be racier, so isalive() can still report True immediately after the
termination signal was sent — a false "could not terminate", not an actual
leaked process.

This kernel's own _cleanup() (registered via atexit) already tolerates
any exception from repl.terminate(), so kernel shutdown itself is
unaffected. But any caller that terminates the REPL directly — e.g.
oct2py's Oct2Py.exit()/restart(), which unregister _cleanup from
atexit and call self._engine.repl.terminate() themselves — sees the
exception propagate.

Changes

• Widen delayafterclose/delayafterterminate on the pexpect child to
  0.5s in OctaveEngine._create_repl(), giving the kernel more headroom
  before isalive() is checked. This is the same spot that already
  removes the default delaybeforesend delay, so it benefits every
  OctaveEngine consumer rather than requiring each one to reach into
  repl.child itself.
• Add regression tests in tests/test_engine.py::TestCreateRepl asserting
  both delays are set on the constructed REPL's child.

Backwards-incompatible changes

None. delayafterclose/delayafterterminate only affect the tail end of
close()'s wait-then-check sequence during teardown; going from 0.1s to
0.5s in the (rare) case where the process is still shutting down does not
change behavior for any caller not currently affected by this race.

Testing

• poetry run pytest: 174 passed.
• just typing: no issues.
• just pre-commit: all hooks pass.

AI usage

• Some or all of the content of this PR was generated by AI.
• The human author has carefully reviewed this PR and run this code.
• AI tools and models used: Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.93%. Comparing base (35cf8c0) to head (d6616d7).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #354      +/-   ##
==========================================
+ Coverage   92.90%   92.93%   +0.03%     
==========================================
  Files           6        6              
  Lines         437      439       +2     
==========================================
+ Hits          406      408       +2     
  Misses         31       31              

Flag	Coverage Δ
linux	92.71% <100.00%> (+0.03%)	⬆️
windows	92.93% <100.00%> (+0.03%)	⬆️

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

Files with missing lines	Coverage Δ
octave_kernel/kernel.py	93.94% <100.00%> (+0.03%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.