More guidance in hardware sizing section

## Communication goals (e.g. detailed howto vs orientation)

The current section - https://docs.citusdata.com/en/v7.3/admin_guide/production_sizing.html - gives numbers somewhat out of context. There's more that goes into choosing cluster size.

## How does this work? (devs)

* MT
* Recommendation is that the working set fits in memory
* If you have an idea how much that is, perfect, else
* Want cache hit ratio and index hit ratio to be 95-99%
* Use that as a proxy for data fitting in memory
* Put hardware sizing before shard count, it flows better that way
* RT
* Strictly depends on latency requirements, you want faster results, then scale the hardware
* Try to go with the working set at least (like previously)
* For number of cores, see how long exist queries go (e.g. 30s), and think what you want that to be (2s) and then bump cores (15x)

These are rough guidelines, and sometimes people's poor performance can be corrected in the database design and queries rather than needing to throw lots of hardware at the problem. It's useful for people to talk with us for advice, so we might include a link or widget to do that.