[alexvalentine94]

Per-solid timeout and log file fix

Summary

A single solid stuck inside OpenCASCADE (an endless loop in IntPatch_WLine::ComputeVertexParameters) could hang a whole multiprocess decomposition, because ProcessPoolExecutor cannot stop one worker. Separately, spawned workers wiped analysis/logging.log.

Changes

• Per-solid time limit: new split_solid_timeout decomposition parameter (default 300 s, 0 = no limit). A solid that runs past it has its worker stopped and replaced, and is saved to failed_to_convert with the reason. The remaining solids carry on.
• New pool: run_with_timeout in decomposemultiprocess.py replaces ProcessPoolExecutor in Decomposer.run. It also replaces workers that die (e.g. a segfault), and returns results in input order.
• Shared failure handling: saving a failed solid is moved into save_failed_solid, which both paths use. Behaviour is unchanged.
• Log file fix: a spawned worker imports mcio before parent_process() is set, with sys.argv copied from the main process, so each new worker truncated the log (causing lost lines and NUL bytes). __init__.py now also checks sys.orig_argv for --multiprocessing-fork.
• Fixed missing spaces in the "conversion failed" message.

Not changed

The decomposition logic itself (splitting, booleans, surfaces, void, output writers). The time limit doesn't apply to single-core or debug runs.

Testing

• New tests/unittests/test_decomposemultiprocess.py (9 tests): a stuck task, all workers stuck, a crashed worker, an exception in the task, result order, no limit, and a log-file regression test that fails on the old code.
• Full suite: 361 passed, 16 skipped.
• On a ~3,800 solid model that used to hang, the run now finishes in about 5 minutes, with 4 solids stopped by the limit and a complete log.

Notes

• Branched from feature/ruff_fix, so merge that first.
• The run still stops with "conversion failed" if any solid fails, as before.