[Feature]: add reasoning/thinking capability and effort level in the benchmark logic

[Fango2007]

Current state:

Model profile/snapshot schemas track whether a model supports reasoning (reasoning.thinking: boolean, reasoning.exposed, reasoning.token_budget_configurable) — but this is metadata only, not used during execution.

buildBenchmarkRequestPayload handles only: temperature, top_p, max_tokens, seed, stop, presence_penalty, frequency_penalty, stream. No thinking/reasoning params injected into requests.

Gap: runtime_parameters has no thinking, budget_tokens, reasoning_effort fields, so they can't be set per-benchmark-run today. A benchmark testing a reasoning model would get default thinking behavior (whatever the server
defaults to), with no way to control or vary the effort level.