Don't depends inconditionally on critical-section for portable-atomic

[ogoffart]

This should fix the compilation (link) of the esp-edf build for esp32s2 in the nightly test CI, as that platform doesn't have atomic, and also we don't provide an implementation of the critical-section function from the slint-cpp crate.

Use the unsafe-single-threaded feature instead.

CC #5057

[Song-aff]

Just wanted to check in—could you kindly update me on the progress of the submission merge?

[ogoffart]

This fails CI in several jobs because

 error: `portable_atomic_unsafe_assume_single_core` cfg (`unsafe-assume-single-core` feature) is not compatible with target that supports atomic CAS;
       see also <https://.com/taiki-e/portable-atomic/issues/148> for troubleshooting
   --> /Users/runner/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/portable-atomic-1.11.0/src/lib.rs:355:1
    |
355 | / compile_error!(
356 | |     "`portable_atomic_unsafe_assume_single_core` cfg (`unsafe-assume-single-core` feature) \
357 | |      is not compatible with target that supports atomic CAS;\n\
358 | |      see also <https://.com/taiki-e/portable-atomic/issues/148> for troubleshooting"
359 | | );
    | |_^

As we enable the Slint's unsafe-single-threaded via the --all-features in our CI. And also for some MCU platform that have atomic CAS.

So this is not a solution. Looks like we'll have to enable the critical-section feature in the compat feature enabled by default.