Consider using `critical_section` for the init-context CriticalSection token

[urdh]

Currently (since rtic-rs/rfcs#32) the init::Context contains a bare_metal critical section token to indicate that the initialization routine runs with interrupts disabled. This causes issues when trying to use RTIC with the STM32F0 HAL and an updated cortex-m crate, which has moved away from using bare_metal to instead use critical_section in rust-embedded/cortex-m#447.

With stm32-rs/stm32f0xx-hal#180, the STM32F0 HAL is going to be usable with more recent cortex-m versions, but this will instead introduce an incompatibility with RTIC which will require users to work around this incompatibility with a bit of unsafe:

    #[init]
    fn init(mut cx: init::Context) -> (Shared, Local) {
        let cs = unsafe { critical_section::CriticalSection::new() };
        // ...
    }

This isn't a show-stopper but given that critical_section::CriticalSection already seems to be used elsewhere in RTIC, and that STM32F0 seems to be the origin of the bare_metal CS to begin with, it seems reasonable to move that CS to critical_section as well.