fix(fileutils): drain pipe buffers after process exits

[steps-re]

Problem

fileutils.run() uses O_NONBLOCK + read_async() to avoid the classic pipe deadlock (write-end full while parent blocks reading the wrong fd). However, when the subprocess exits run() breaks out of the read loop as soon as p.poll() returns non-None — potentially before all data the process wrote has been drained from the kernel pipe buffer.

With O_NONBLOCK a single read() syscall returns only what the kernel has ready at that instant; a large burst of output written just before process exit (common in nuclear simulation and optimization codes) can silently disappear. This is the underlying cause of the hang/truncation behaviour reported in #49.

Fix

Add a post-exit drain loop for both stdout and stderr (2 + the while True loop in fileutils.py). Once the write end of the pipe is closed, read_async() returns b"" (EOF) rather than EAGAIN, so the loop terminates immediately after all buffered bytes are consumed.

Test

Added test_run_captures_all_output which emits 160 000 bytes of output — well past a typical 64 KB pipe-buffer read — and asserts every byte is captured. The test is skipif win32 (same as the rest of the non-blocking path).

tests/test_fileutils.py::test_run_captures_all_output PASSED

All 4 test_fileutils tests pass.

Fixes #49

---

Signed-off-by: Mike German mike@stepsventures.com