Add process stop cleanup regression coverage

[johnynek]

Adds cross-backend regression coverage for the portable process stop/status contract and managed with_process cleanup behavior. The shared JVM/Python process test now exercises stable status observations, timeout non-consumption, already-exited stop behavior, terminate/kill post-stop stability, low-level stdio ownership through explicit close, and helper-owned versus caller-owned handle cleanup. C/libuv tests add matching low-level coverage for stable recorded status and ensure poll, zero-duration wait_timeout, terminate, and kill do not implicitly close returned pipe handles.

Fixes #2376

[gemini-code-assist]

Code Review

This pull request introduces comprehensive regression coverage for the portable process stop and status contract across the C/libuv, JVM, and Python backends. Key changes include the addition of test helpers to verify that process exit status remains stable across repeated observations and new test cases ensuring that low-level operations like poll, terminate, and kill do not implicitly close caller-owned stdio pipe handles. Feedback suggests strengthening these ownership tests by using non-empty payloads during write operations, as zero-byte writes might be optimized away by certain runtime implementations, potentially failing to detect if a handle was incorrectly closed.

[gemini-code-assist]

Writing an empty string to stdin_h may be a no-op in some backend implementations and might not actually verify that the handle remains open and valid. To robustly prove that the low-level stop functions did not close the returned handles, consider writing a non-empty string. Note that a flush operation should be avoided for this purpose as it is intended to be a no-op for performance on certain file descriptors.

          case Some(h): write_utf8(h, " ")

References

1. For file descriptors without user-space buffers (e.g., libuv's uv_file), a flush operation should be a no-op for performance.

[gemini-code-assist]

Similar to the Bosatsu test case, writing zero bytes (io_core_bytes_value(NULL, 0)) might be optimized away by the runtime or the underlying libuv implementation, potentially failing to detect if the handle was incorrectly closed. Using a non-empty payload, as done in the terminate test case at line 1373, would provide a stronger regression check for handle ownership.

      ___bsts_g_Bosatsu_l_IO_l_Core_l_write__bytes(slots[1], io_core_bytes_value((const uint8_t*)" ", 1)),

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.21%. Comparing base (a4aa784) to head (919c80f).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2377      +/-   ##
==========================================
+ Coverage   85.18%   85.21%   +0.03%     
==========================================
  Files         196      196              
  Lines       49379    49379              
  Branches    12196    12196              
==========================================
+ Hits        42062    42078      +16     
+ Misses       7317     7301      -16     

☔ 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.

[johnynek]

Blocked before ready-to-push: I made the requested non-empty write changes and uncovered a JVM ownership issue where terminate via java.lang.Process.destroy() closes the child stdin stream. I patched JVM terminate to use ProcessHandle.destroy() and changed the Bosatsu ownership fixture to prove a non-empty post-terminate write before killing the child for cleanup; the C kill-path test now writes a one-byte payload. However, a required-tests run started before the final cleanup patch hung in the sandbox, and this environment does not permit process listing or pkill, so I could not obtain a clean passing scripts/test_basic.sh result in this turn. Because the required gate did not complete, this is not ready to push.