Remove bottleneck in reconciliation controller

[jveski]

It turns out the default controller work queue rate limiter has a shared token bucket of 10 RPS (with some reasonable burst) that applies to all requeues. Since the reconciliation controller can't/doesn't watch the resources it writes, it relies on requeues to form its feedback loop, which means when the loop ticks into the second iteration of any resource (typically the status update) it will wait a few dozen milliseconds.

This also fixes a couple of other small nits surrounding client-side rate limiting.