openkal-musl 0.6.0: a caller's redirection reaches the program it starts

[Sunrisepeak]

openkal-musl 0.6.0, released and mirrored. openkal-llvm-runtime 0.3.1
follows in its own change once its own run on main finishes; it records which
C library that version is handed and adds nothing else.

What 0.6.0 answers

Three defects a consumer reported as
openkal-linux#13. Each is
a place where the specification already carried the atom and the C library had
no route to it; openkal itself does not change.

	what a program observed	what carries it now
a caller's redirection	dup2(fd, 1) and then posix_spawn, system, or fork + execve started a program writing to the stream the caller had been started with; the file it had redirected onto stayed empty	the three positions are seeded from the descriptor table
abort	it did not end the program. musl's raise is tkill, the port had no case for it, so abort fell through to a_crash() — hlt on x86_64 — and every uncaught C++ exception, assert and std::terminate reported a segmentation fault	SYS_tkill/SYS_tgkill perform a signal's default action; SIGABRT reaches kal_abort
waitpid(…, WNOHANG)	it blocked	kal_timeout_wait_process

The second row is openkal-llvm-runtime's business directly: libc++abi's
terminate handler ends in abort, so every uncaught C++ exception over that
runtime reported a segmentation fault.

Verification

• openkal-musl: four rows (Linux gcc, Linux llvm, macOS llvm, Windows gcc)
  green on main; examples/subprocess grew from 7 observations to 24.
• openkal-llvm-runtime: five jobs green, built against openkal-musl's branch
  rather than a published version — including the bare-metal cfg(os = "none")
  configuration, which is the only place that compiles the C library with
  OKM_HAS_FS=0 and OKM_HAS_PROCESS=0.
• Measured against the previous release with the new probe: -- failures: 7 --,
  one for each fix, with fifteen controls holding.
• The published pair, resolved from this index in a sandbox that has none of
  these packages installed, builds and runs a program asserting all three.

The digest is of the archive GitHub serves, and the CN asset is that same file:
downloaded from the mirror and compared, ca769663… both ways, 990799 bytes.