improve exception handling in CompletionStage* strategies

[Ladicek]

All CompletionStage* strategies currently have duplicated exception handling, both for synchronous case and asynchronous case. They typically look similar to:

try {
    CompletableFuture<V> result = new CompletableFuture<>();
    delegate.apply(ctx).whenComplete((value, exception) -> {
        if (exception == null) {
            result.complete(value);
        } else {
            result.completeExceptionally(exception); // <1>
        }
    });
    return result;
} catch (Exception e) {
    return failedStage(e); // <2>
}

This is just the core, all the strategies do metric updates, internal state updates, strategy-specific logic, etc. This all has to be duplicated between <1> and <2>.

It would be much better if we could just rely on asynchronous exception handling. The only place exceptions can be thrown synchronously from is (or should be) the ultimate end-user method call (i.e., the Invocation strategy). We can wrap it into another strategy that will make sure exceptions are not propagated synchronously. Implementation of such strategy is actually the exact code given above (but we'd use CompletionStages.propagateCompletion instead). All the other CompletionStage* strategies then don't have to care about synchronous exceptions and can drop the try/catch structure.

[Ladicek]

At this point, I'm not exactly sure we should do this. The synchronous exception handling at least guards against mistakes in our own code -- which are really easy to make, because exceptions can be thrown from anywhere. I'll keep this issue open for now, because the duplication isn't nice and there might be something else we could do.