std::fmt::Debug for Layer implementations

[koivunej]

Follow-up to #3513.

This removes the old blanket std::fmt::Debug impl on dyn Layer which did not seem to be used from anywhere (no compilation errors after removing).

Adds std::fmt::Debug requirement and implementations for trait Layer implementors:

• LayerDescriptor (derived)
• RemoteLayer (manual)
• DeltaLayer (manual)
• ImageLayer (manual)

Adds and adjusts some doc comments to be more rustdoc alike.

In follow-up #3544: Use this std::fmt::Debug when logging a LayerMap::replace failure for which we currently have no tests but speculation (see #3387 comments).

Alternatives for adding the std::fmt::Debug:

• add a custom Layer::get_type_name(&self) -> &'static str
  • would give no additional information to the place of use except the type name
• add other custom std::fmt::Debug alike functionality to the trait
  • std::fmt::Debug is the way to debug format objects

[problame]

Nice job!

[koivunej]

This should probably wait eviction related stats because we might want to include those.

[koivunej]

After looking a bit how to use this on the new evict_layer endpoint, I am thinking I'll just separate that to different PR altogether (with the c8b3139).

[SomeoneToIgnore]

I'm a bit sad that it makes it harder (at least, a bigger patch) to go away from the "&dyn Trait as an interface" approach, but otherwise looks good.

[koivunej]

@SomeoneToIgnore please elaborate how will this make it more difficult? I think the main point is that we should have Debug for Layer impls, surely this is as easy to do with more typed values than not?

[SomeoneToIgnore]

Maybe I'm wrong, just imagined that now every place with Arc<dyn Layer> could be debugged, and for the later swap, more code changes could be required in either of such place.
But it's still noting big, having Debug seems to be a good thing in general.

[koivunej]

more code changes could be required in either of such place.

Oki now I understood. I think in future it's just important to use #[derive(Debug)] regardless of the shape. Here it cannot yet be used because we could dump a lot of indices for InMemoryLayer and few of those extra fields, perhaps EphmeralFile would had had quite long Debug output as well (the full hashmap of files). EDIT: Forgot there were the tenantid and timelineid as well, and conf. They are everywhere :)

I don't want to drop fields (above discussion on RemoteLayer I'm keeping open) or add newtypes which would suppress any long output (InMemoryLayer's index) to move towards that on this PR but having Debug impls in general will be a good thing.