Reorder parsing of type qualifiers to match the spec.

[khuey]

The spec says that the qualified-type production (within type) is parsed second only to the builtin-type production. The qualified-type production is ambiguous with later productions (in particular with the "Ul" inside class-enum-type) so the order does matter. This reorders things for "Ul" and brings along the other parts of the qualified-type production.

To preserve the existing behavior for function-type, rather than attempt to pass down a leading CV-qualifiers we simply skip them if the tail remaining indicates a function-type production is imminent.

This is the root cause of @Swatinem's issue, though perhaps some defense in depth that limits runaway growth of substitution tables is still wise.

[Swatinem]

This indeed covers the pathological case. Here is a test that now runs instantly, whereas it would previously spin for ~15 seconds for production builds. Feel free to also add it to the testsuite.

// This symbol previously ran into some mutual recursion and unbounded growth of the substitution table.
// See <https://github.com/gimli-rs/cpp_demangle/issues/277> and <https://github.com/getsentry/symbolic/issues/477>
#[test]
fn test_pathological_recursion() {
    let s = "_ZUlzjjlZZL1zStUlSt7j_Z3kjIIjIjL1vfIIEEEjzjjfjzSt7j_Z3kjIIjfjzL4t3kjIIjfjtUlSt7j_Z3kjIIjIjL1vfIIEEEjzjjfjzSt7j_Z3kjIIjfjzL4t3kjIIjfjzL4t7IjIjjzjjzSt7j_Z3kjIIjfjzStfjzSt7j_ZA3kjIIjIjL1vfIIEEEjzjjfjzSt7j_Z3kjIIjIjL1vfIIEEEjzjjfjzSt7j_Z3kjIIjfjzL4t3kjIIjzL4t7IjIjjzjjzSt7j_Z3kjIIjfjzStfjzSt7j_ZA3kjIIjIjL1vfIIEEEjzjjfjzSt7j_Z3kjIIjIjL1vfIIEEEjzjjfjzSt7j_Z3kjIIjfjzL4t3kjIIjfjzL4t7IjIjL1vfIIEEEjzjjSI";
    let parse_options = cpp_demangle::ParseOptions::default().recursion_limit(160); // default is 96
    if let Ok(sym) = cpp_demangle::Symbol::new_with_options(s, &parse_options) {
        panic!("Unexpectedly parsed '{}' as '{}'", s, sym);
    }
}