[enhancement] Add user-controlled "target throughput"

### What's hard to do? (limit 100 words)

Automatically reuse resources between pipeline stages (with appropriate guards) if the user wants to design a target that does not need to achieve full throughput. For example, if the user wants an II=2 circuit (best-case throughput = worst-case throughput = 2), we should be able to share resources (e.g., pipeline registers) between adjacent stages.

### Current best alternative workaround (limit 100 words)

The user can implement a manual finite-state machine, overriding XLS's normal pipelining to various degrees in order to force more sharing.

### Your view of the "best case XLS enhancement" (limit 100 words)

Add a `--target_throughput` flag to XLS scheduling, much like the current `--worst_case_throughput` flag. Once this is in place, we can start adding logic allowing adjacent stages to share common resources.