Add option for skipping fields

[pnevyk]

Implements #72

The idea is similar to serde's skip. The getter/setter is not implemented for fields that are marked #[getset(skip)]. So for:

#[derive(CopyGetters)]
#[get_copy]
pub struct Plain {
    #[getset(skip)]
    non_copyable: String,
    copyable: usize,
}

only fn copyable(&self) -> usize getter is generated. This feature makes sense for larger structs, where using struct-level attribute is convenient for brevity, but the user still wants to ignore a subset of fields. It is possible to achieve this behavior in the current version by annotating each field with a corresponding getset attribute and not annotating the fields for which the getter/setter should not be generated.

This PR lacks thorough testing and documentation, I wanted to be sure this is a way to go before writing the tests and docs.

Open questions I can think of:

1. Current implementation accepts multiple identifiers inside getset attribute. For the example above, it means that if I use #[getset(get_copy, skip)], it silently swallows the "get_copy" part and skips the field, but if I use #[getset(skip, get_copy)], the last identifier (i.e., "get_copy") is taken and the getter is generated (resulting into a compilation error for moving a non-copyable value). This is inconsistent and confusing for the user. I think it should be forbidden to use any other identifier if skip is present, but there is also point 2.
2. A user might want to skip only a particular getter/setter. For example, the struct could derive both Getters and Setters and the user wants to omit the setter part on a field. Should getset support this too by a more granular syntax (e.g., getset(skip(get)) or get(skip) or getset(skip_get)) or is it out of scope? (The getset(skip_get) syntax is inspired by serde's skip_serializing.)

[pnevyk]

The code now checks whether some getter or setter is used with skip and gives a compilation error if that is the case. I believe this is the least confusing behavior as it disallows the user to use a getter/setter with skip, possibly thinking that it somehow influences the generation. This made the implementation of parse_attr a bit more complicated and less nice. Maybe there is a better way to do it?

I leave the "skip a specific getter/setter" feature out of the PR.

Any ideas how this can be tested further?

[Hoverbear]

This looks good, thanks @pnevyk :)

[pnevyk]

Great, thank you too