Improve explanation of not_unsafe_ptr_arg_deref

[dtolnay]

I pointed out in #7666 (comment) that the use of allow(clippy::not_unsafe_ptr_arg_deref) on crates.io is shockingly high.

Here is the list: https://dtolnay.github.io/noisy-clippy/not_unsafe_ptr_arg_deref.html

I spot checked some of these and pretty much every case is unsound.

• Public API of bp7 exposes UB to safe code dtn7/bp7-rs#2
• Public API of aeron-rs exposes UB to safe code UnitedTraders/aeron-rs#14
• AtomicBuffer still exposes UB to safe code UnitedTraders/aeron-rs#16
• dealloc_buffer_aligned leads to UB in safe code UnitedTraders/aeron-rs#17
• Public API of drogue-ffi-compat exposes UB to safe code drogue-iot/drogue-ffi-compat#1
• Undefined behavior in release_string_utf_chars jni-rs/jni-rs#331
• Public API of mosquitto-jwt-auth exposes UB to safe code wiomoc/mosquitto-jwt-auth#11
• Public API of pact_ffi exposes UB to safe code pact-foundation/pact-reference#148
• Public API of spirv-tools exposes UB to safe code EmbarkStudios/spirv-tools-rs#27
• Public API of tm-rs exposes UB to safe code JoNil/tm-rs#1
• Public API of pgx exposes UB to safe code pgcentralfoundation/pgrx#196
• is_a is unsound pgcentralfoundation/pgrx#215

If people are not understanding what this lint is attempting to tell them, it seems like the lint isn't doing its job. We should look into whether the lint's documentation or the lint's diagnostic output can be improved in a way that better comes across to people. It may be worth looping in some of the maintainers of some of the crates above about what might have been helpful for them to understand the lint.

[tgross35]

Opened #10116 with stronger wording & context

Has uplifting this lint to rustc ever been considered? It seems important enough to be considered (I'm not sure what the criteria are for that)