Automate CircleCI configuration generation using Jinja2 templates

[divya0709]

Summary

This PR introduces an automated workflow for generating the .circleci/config.yml file.

Details

• Added .circleci/config.yml.in : a Jinja2-based template defining reusable structure.
• Added .circleci/regenerate.py : Python script that renders config.yml automatically using version matrices, CUDA configurations, and branch filters.
• Generated .circleci/config.yml from the above automation.

Benefits

• Reduces manual maintenance of CircleCI config.
• Ensures consistent YAML formatting and up-to-date workflow definitions.
• Easier updates when adding/removing Python, CUDA, or OS combinations.

Notes

• The unittest/ directory remains unchanged.
• The macOS Conda Python 3.9 job was commented out in the original config.yml, so it is intentionally skipped in this generated configuration for consistency

Fixes #97

[divya0709]

Note: All CircleCI jobs fail in my fork due to missing Docker resource classes (2xlarge+, gpu.medium).
This is expected since contributor forks don’t have access to Meta’s internal CircleCI plan.
YAML has been validated locally : config structure should be correct.