Proposal : Benchmarking Improvement - Aligning measure_runtime with time_fn in time_utils.py #4747

☑️ I understand it is strictly prohibited to use AI to write issues.

I want to improve the shared python benchmark timing helper by making measure_runtime sync the returned MLX values like in time_fn

If a callable returns an unevaluated MLX value, timing without mx.eval could measure host-side dispatch rather than the actual completed device work

Configurable warmup and iteration counts would also improve benchmarking by making it a little more modular - I will preserve the current defaults

Please let me know whether this scope would be useful, or if you would prefer a different approach before I begin implementation