return HTTP429 on dispatch if eval broker queue depth is too high

For clusters that primarily serve dispatch workloads, it could be useful to provide backpressure when the cluster has a long backlog of pending evaluations. For backwards compatibility we'd want this feature to be disabled by default, as well as configurable by the cluster administrator.

(Internal ref https://hashicorp.atlassian.net/browse/NMD-948. See also internal ref https://hashicorp.atlassian.net/browse/NMD-945 for an RFC I'm working on for eval broker concurrency controls and multi-tenancy.)

Start by reviewing the issue's proposed backpressure behavior and the referenced NMD-945 RFC for eval broker concurrency controls and multi-tenancy. The work is done when dispatch requests return HTTP 429 at a configurable queue-depth threshold, while the feature remains disabled by default and administrator-configurable.

Written by the indexing model from the issue text.