verify arguments of manual benchmark and align with scripts

[jfomhover]

Verified the manual benchmark docs page and fixed the following:

• made device_type optional when running local
• explained device_type in the docs
• added num_thread as it could be relevant for lightgbm inferencing
• removed the lightgbm_cli section as we're trying to deprecate this (see #165

[github-actions[bot]]

Unit Test Results for Build

  1 files  ±0    1 suites  ±0   6s :stopwatch: ±0s 75 tests ±0  75 :heavy_check_mark: ±0  0 :zzz: ±0  0 :x: ±0 

Results for commit b7db510e. ± Comparison against base commit a5b21bc7.