Identify the limits of each component and a cluster as a whole

We want to answer the following questions better:
- [ ] How many instances each role can have?
- [ ] How much performance can we get from each component?
- [ ] To get x% more performance from a component (say proxy), how many new instances and how much more resources do we need to add? Is there a mathematical model that calculates a component's performance based on the component's allocated resources?
- [ ] What are the maximum shard counts, data size a cluster can hold?

[This idea was proposed and discussed in FDB Offset 2019.]

No files, tests, or entry points are identified. Start by reviewing the FDB Offset 2019 discussion referenced in the issue, then define how each checklist question would be measured for components and a cluster. Done would require documented limits, performance/resource models, and maximum shard and data-size findings.

Written by the indexing model from the issue text.