Missing semihosting printouts after upgrade from v0.3.3 to v0.5.0

In cortex-m-rtic CI cortex-m-semihosting is used for printouts and QEMU control.

As part of this PR#680 semihosting got bumped to latest v0.5.0, and after adopting code to match the removed error handling, CI still failed with the following:

👟 cargo run --example binds --target thumbv7m-none-eabi --features test-critical-section --release
   Compiling cortex-m-rtic v1.1.3 (/home/henrik/dev/rust/rtic/cortex-m-rtic)
    Finished release [optimized] target(s) in 0.17s
     Running `qemu-system-arm -cpu cortex-m3 -machine lm3s6965evb -nographic -semihosting-config enable=on,target=native -kernel target/thumbv7m-none-eabi/release/examples/binds`
Timer with period zero, disabling
init
foo called 1 time
idle

Error: Differing output in files.

Expected:
init
foo called 1 time
idle
foo called 2 times


Got:
init
foo called 1 time
idle

which given that the idle task in examples/binds.rs consists of:

    #[idle]
    fn idle(_: idle::Context) -> ! {
        hprintln!("idle");

        rtic::pend(Interrupt::UART0);

        debug::exit(debug::EXIT_SUCCESS); // Exit QEMU simulator

        loop {
            cortex_m::asm::nop();
        }
    }

suggests that the rtic::pend(...) in the idle task did not get to run.

Modifying the code by placing the debug::exit(...) after the NOP inside the loop,
then all expected printouts gets received by the host.

Which modifies the hypothesis to "printout did not reach host before QEMU exited"

See this commit for the "fix" to the CI failures: 9764121c

Focusing on the binds example, this is an objdump diff together with the diff, however, not built with --release:

Another more complete picture, all built with release, the splits starting from the left:

• cortex-m-semihosting v0.3.3 idle task
• v0.5 without the "fix" idle task
• v0.5 with the "fix" idle task

Generating the objdumps with

cargo objdump --example binds --target thumbv7m-none-eabi --release -- -d --no-show-raw-insn --print-imm-hex 

Curious to hear if this is something seen before, or if a possible fix would be to always throw in a NOP as part of the syscall when debug::exit-iting.

All tests were run with latest stable rustc 1.66.1 (90743e729 2023-01-10)

I have not tried different versions of QEMU, but it could be something on that end too...

Start with examples/binds.rs and the cortex-m-semihosting v0.5.0 upgrade described in PR#680. Reproduce the release build with cargo run for the thumbv7m-none-eabi target under QEMU, then compare objdumps using the command in the report. Done means the expected second printout is received before QEMU exits without relying on moving debug::exit into the loop.