refactor: use time.perf_counter for consistent and accurate benchmarking

[Satyam12singh]

Proposed Changes:
This PR updates the timing logic in several benchmark scripts to use time.perf_counter() instead of time.time().

Files modified:

• benchmarks/python/comparative/bench_mlx.py
• benchmarks/python/comparative/bench_torch.py
• benchmarks/python/time_utils.py
• docs/src/dev/extensions.rst
• examples/python/linear_regression.py

Why this is necessary:

1. Monotonicity: time.time() can be affected by system clock updates (NTP sync), whereas time.perf_counter() is a monotonic clock that never goes backward.
2. Precision: High-performance kernels (like those in MLX and Torch) require the highest resolution possible to calculate accurate "msec" per iteration.
3. Consistency: This aligns the benchmarking utility functions with the style already adopted in the rest of the MLX examples.

---

Checklist

• I have read the [CONTRIBUTING](https://github.com/ml-explore/mlx/blob/main/CONTRIBUTING.md) document
• I have run pre-commit run --all-files to format my code / installed pre-commit prior to committing changes
• I have added tests that prove my fix is effective or that my feature works
• I have updated the necessary documentation (if needed)