Add standardized benchmark suite for interface overhead and sampling audit

[saksham-stack]

This PR introduces a dedicated benchmarking suite to the dingo repository. As discussed in volesti#423, the goal is to provide a repeatable way to measure the performance overhead introduced by the Python interface and to audit the numerical consistency (convergence) of the sampling process.

By moving this logic into a standardized /benchmarks directory, we ensure that future changes to the C++ core or Python wrappers can be audited for performance regressions.

Changes
• Added /benchmarks directory: Established a new package for performance-related tests.

• test_samplers_bench.py: Implemented a pytest-benchmark suite that measures the wall-clock time for sampling from high-dimensional polytopes (e.g., unit cube).

• metrics_utils.py: Integrated ArviZ to calculate Effective Sample Size (ESS), addressing the need for automated convergence diagnostics.

•Documentation: Added a README.md and requirements-bench.txt to guide contributors on how to run and interpret these benchmarks.

Verification Results
Running pytest benchmarks/ locally on [Your OS/Processor] yielded the following for a 10D unit cube:

• Mean Sampling Time: [Insert your result from terminal, e.g., 12.5ms]

• Avg ESS: [Insert your ESS result, e.g., 450.2]

•Interface Overhead: [Insert your calculated overhead %]

GSoC 2026 Context
I am a prospective GSoC 2026 contributor. This task has helped me familiarize myself with the dingo sampler interface and the integration of volesti bindings. I plan to expand this suite to include more complex geometries (e.g., simplices) as the project progresses.