shell: fail the command when a > ${buf} redirect overflows the target Buffer - #34698

The `"<cmd>: No space left on device"` diagnostic is only emitted when `stderr.needs_io().is_none()`, so in the default (non-`.quiet()`) case — where stderr is `BuiltinIO::Fd` teeing to the terminal + captured buffer — the message is silently dropped and the command just exits 1 with nothing on stderr and nothing in `r.stderr`. All the added tests use `.quiet()`, so this path is uncovered. If skipping the async enqueue here is deliberate (to keep `done()` synchronous), it's worth a code comment;

Two minor issues with this branch: (1) it clobbers the subprocess's own nonzero exit code — `buffered_output_close_stdout`/`_stderr` (Cmd.rs:988-990/1024-1026) unconditionally set `exit_code = errno`, so a subprocess that both overflows the buffer *and* exits nonzero has its real code replaced by 28 (asymmetric with the builtin path's `(code, _) => code` guard); and (2) `to_system_error()` heap-allocates a `WTFStringImpl` message that the sink never `.deref()`s, leaking it per overflow. Swapping

The doc comment on `Builtin::write_no_io` still reads "Returns `Err(ENOSPC)` when an ArrayBuffer target is already full", but after this change `write_no_io_to` returns `Err(ENOSPC)` on *any* short write into an ArrayBuffer (after writing what fits), not just when it was full on entry. Suggest updating to something like "Returns `Err(ENOSPC)` when an ArrayBuffer target cannot hold all of `buf` (after writing what fits)."