Clarify Recommendation Mapping Logic (Traffic Profile & SLO Levels)

[toslali-ibm]

It would be nice to have a clear documentation explaining how the specification mapping logic works—specifically how traffic profiles and SLO levels are interpreted and mapped.

• What do the High / Medium / Low levels actually represent?
• How are these levels translated into: expected request rates, concurrency, latency/SLO requirements, throughput assumptions?

[jgchn]

Connects to llm-d/llm-d-benchmark#482