Worker manager assignment strategy

Issue Description
===
Worker manager should assign server instances according to the entire execution graph, not blindly assigning instances and hope the distributed plan generator figures things out later; currently we blindly assign every server on the tenant for intermediary stages.

Design
===
Here is the design doc for the global worker assignment strategy V2
- [Worker Assignment Design Doc](https://docs.google.com/document/d/1SKDKV6LXr4uFFUsR3djz5BWWMqcSJIYEqJBoL1zeDD8/edit)
- [Partition Design Doc](https://docs.google.com/document/d/1CdvxmOOctk6kS5PdgCy7f5KVh5urw4YY0YZGbwuPJt4/edit#)

TODO
===
- Physical
- [ ] Refactor post-logical planner into single physical planner https://github.com/apache/pinot/pull/10481
- [ ] Refactor dispatch to batch dispatch based on physical server + global partition set https://github.com/apache/pinot/pull/10493
- [ ] Create new worker assignment function
- Logical
- [ ] rework calcite to allow it to construct RelTrait-aware RelNode that contains RelDistribution
- [ ] allow colocation & other optimization hints
- Partition
- [ ] partition aware routing: https://github.com/apache/pinot/pull/10461

Start with the Worker Assignment Design Doc and Partition Design Doc, then review the linked pull requests #10481, #10493, and #10461 to understand the planned physical, dispatch, and partition changes. Done means the listed physical, logical, and partition TODOs support global worker assignment across the entire execution graph.

Written by the indexing model from the issue text.