Generate a struct for each operation's variables

[saga-dasgupta]

Based on #147, so the first commit here is #147's. This PR is the second commit, which also adds a #147 test case overriding an input field to a variables struct. I'll rebase once #147 merges.

Typegen generates types for what an operation returns, but not for the variables it declares. A caller building those variables by hand gets no help from the compiler: renaming, adding, removing or retyping a variable in the query still compiles, and only fails once the query runs.

This generates a struct for an operation's variables, next to the operation's own struct in the query module:

query MyArgs($required: String!, $optional: Int, $withDefault: Int! = 1) {
  myArgs(required: $required, optional: $optional, withDefault: $withDefault)
}

pub struct MyArgsVariables {
    pub required: String,
    pub optional: Option<i32>,
    pub with_default: Option<i32>,
}

• Naming: <Operation>Variables, or RootVariables for an anonymous operation, whose struct is Root. Operations that declare no variables get no struct.
• Fields: a variable that is non-null with no default is a plain field. Any other variable is an Option, so a variable with a default can be left out.
• Field types: the same as input object fields. The type-building code moves out of input_object_type_definition.rs into a new input_type.rs that both use. That includes borrowing: with borrow = true, the struct gets an 'a lifetime when a field borrows.
• CodeGenerator hooks: attributes_for_variables_struct, additional_impls_for_variables_struct and attributes_for_variables_struct_field, modelled on the input object hooks. All three default to no-ops, so existing implementations don't change.
• bluejay-typegen-macro:
  • derives Clone, PartialEq, Debug and serde's Serialize;
  • renames each field to its GraphQL variable name;
  • skips an optional variable that is None, so the query's default applies instead of an explicit null.
• Validation: it's a typegen error when a variables struct's name clashes with the struct for a fragment or another operation in the same module, for example query My($x: Int) next to fragment MyVariables on .... Without this check, rustc would report a duplicate definition.

Compatibility

These are cases that compile on main and don't here:

• A document with an operation that declares variables, alongside a fragment or operation whose struct already has that <Operation>Variables name. This now gets the clash error above.
• A variable named $self, $Self, $super, $crate or $_ now panics in names::to_ident. These names can't be raw identifiers, so format_ident!("r#self") panics. Fields and input fields with those names already panic the same way on main; variables just didn't become identifiers before. I left this alone to keep the PR focused, but I can add a typegen error for it here if you'd prefer.
• Two variables whose names map to the same field, such as $fooBar and $foo_bar, now fail with rustc's duplicate field error (E0124). Input objects with such fields already fail the same way.
• With bluejay-typegen, a custom scalar that is a variable's type, but no input object field's, now needs Clone, PartialEq, Debug and serde's Serialize, which the variables struct derives. Input object fields already need these.

[saga-dasgupta]

Moved to #149, stacked on #148's branch so its diff is only this change.