Improve object keys performance

[joshkel]

I noticed that constructing Joi.object schemas of complex object types was causing significant slowdown in my Node.js application startup. Profiling revealed that a major source of slowdown was the topographical sorting within object keys' rebuild: sort was called for each added property, resulting in N * N topographical sorts.

To address this, this PR uses @hapi/topo's manual option to sort once, at the end. If that sort fails, it redoes the rebuild operation, sorting per key as the current behavior, so that tryWithPath can throw the more detailed error message.

The updated code constructs my application's schemas roughly 3.6x faster.

The preexisting test suite did not cover rebuild / @hapi/topo's error handling, so I added tests to cover the affected behavior.

[Marsup]

Very nice find! Though I'd go for a less repetitive version like calling an external function:

schema.$_terms.keys = new internals.Keys(...internals.sortKeys(schema.$_terms.keys));

.....

internals.sortKeys = function (keys, manual = true) {

    const topo = new Topo.Sorter();
    for (const child of keys) {
        Common.tryWithPath(() => topo.add(child, { after: child.schema.$_rootReferences(), group: child.key, manual }), child.key);
    }

    try {
        return topo.sort();
    }
    catch {
        return internals.sortKeys(keys, false);
    }
};