Document recommended benchmarking practice

[heshamelmatary]

Having a set of guidelines while benchmarking CheriBSD will be good. Currently I am aware that we should:

• Build NODEBUG kernels (which disables INVARIANTS and WITNESS).
• [malloc] Build with MALLOC_PRODUCTION

[jrtc27]

We ship kernels built with NODEBUG, and ship libc built with MALLOC_PRODUCTION. Instructions for users should not have any mention of rebuilding CheriBSD, and if we’re missing something they need we should ship it so they don’t have to.

[rwatson]

However, it would be good to state that ..

• State that benchmarking on a prototype microarchitecture and with a research compilers and OSes is challenging, as experimental design can be complex, and the data can be difficult to interpret
• 23.10 is the first release suitable for generalised benchmarking as:
  • the benchmark ABI exists, should be used, and cross reference to the performance report
  • third-party packages in prior releases were not compiled with optimisation, and included debug symbols
• If there are experiment design questions, or help is needed in interpreting results, the CHERI-CPU Slack is a good place to ask