fix(spawnSync): throw instead of segfaulting when event-loop creation fails - #35033

On Windows, this new early return leaks any `Box<uv::Pipe>` allocations sitting in `extra_fds` (populated by `as_spawn_option()` for stdio slots ≥ 3) — `WindowsStdio` deliberately has no `Drop`, and `spawn_options.deinit()` (which would iterate `extra_fds` and call `.deinit()`) hasn't been reached yet. This is heap-only (~`sizeof(uv_pipe_t)` per slot, never `uv_pipe_init`'d so no OS handle) and the same pattern already exists on several pre-existing early returns in this function, so it needn't 

On the `mach_port_set_attributes` failure path, `mach_port_mod_refs(..., MACH_PORT_RIGHT_RECEIVE, -1)` destroys the port but the send right added by `mach_port_insert_right` is converted to a dead-name entry that persists in the task's port name table (the code comment even says so). Add `mach_port_deallocate(self, cb->port);` after the `mod_refs` call to release it — same as `us_internal_async_close` does for the send right on the normal-close path. Essentially-unreachable trigger, but this blo