core: Warn about using irq_enable

[chrysn]

Contribution description

Setting the system in a state with more interrupts enabled than expected by the caller is generally unexpected -- if this were allowed unchecked, many components would need to revisit and tighten their synchronization efforts making them less efficient. (I didn't do a precise check, but I expect that ztimer might not tolerate interrupts to be enabled by the callback, especially as that means that unrelated interrupts might manipulate ztimer again).

One offending user was found in the implementations (motor_driver) and changed to use irq_restore. All others have some justification for using enable (eg. it's done in an ISR).

Testing procedure

• Look at whether you agree with the doc change.
• Test the motor example. (I didn't, and don't have the hardware -- but what could go wrong?).

Issues/PRs references

This came up looking at the synchronization necessary around Rust code. If arbitrary callbacks can enable interrupts just like that without responsibility, code in critical sections just needs way more severe locking and precautions.

[edit: since when does GitHub accept submissions with plain Enter? Sucks...]

[kaspar030]

ACK.