For now, only pull in atomic-polyfill for cortex-m feature

It turns out the `AtomicCheckMutex` which was the reason for adding
`atomic-polyfill` is currently gated behing the `cortex-m` feature.
Thus we can gate the dependency behind this feature-flag to resolve
issues on other platforms in the meantime.

In the long run, `AtomicCheckMutex` should get its own feature-flag but
for now this is the easiest solution and a non-breaking change.

Ref: #31, #33

Author: Rahix

Cargo.toml

@@ -22,11 +22,12 @@ once_cell = { version = "1.4.0", optional = true }
 cortex-m = { version = "0.6.3", optional = true }
 xtensa-lx = { version = "0.6.0", optional = true }
 spin = { version = "0.9.2", optional = true }
-atomic-polyfill = "0.1.6"
+atomic-polyfill = { version = "0.1.6", optional = true }
 
 [dev-dependencies]
 embedded-hal-mock = "0.8"
 
 [features]
 std = ["once_cell"]
 xtensa = ["xtensa-lx", "spin"]
+cortex-m = ["dep:cortex-m", "atomic-polyfill"]