Add initial implementation of FCOS to official/projects

[Samir-atra]

Implementation of the FCOS (Fully Convolutional One-Stage Object Detection) model as a new project in the TensorFlow Model Garden.

This PR addresses issue #10275: [Help wanted] FCOS: Fully Convolutional One-Stage Object Detection.

This works corresponds to the request to implement the FCOS paper. The implementation provides an anchor-free object detection framework with the following components:

• Backbone: ResNet50 (configurable via standard TF Model Garden mechanisms).
• FPN: Feature Pyramid Network for multi-scale feature extraction.
• Heads: Shared heads for Classification, Box Regression, and Centerness.
• Losses: IOULoss, FocalCrossEntropy, and BinaryCrossEntropy for centerness.

Type of change

• A new research paper code implementation
• New feature (non-breaking change which adds functionality)

Tests

I have performed local testing to verify the integrity of the model architecture and the training loop.

• Unit Tests: Added official/projects/fcos/model/model_test.py to verify model instantiation and forward pass shapes.
• Integration Test: Ran main.py (training loop) locally with a subset of COCO data to verify the pipeline from data loading to gradient updates.

Test Configuration:

• Hardware: Need computation resources sponsorship for training
• OS: Linux
• TensorFlow Version: 2.x

Checklist

• I have signed the Contributor License Agreement.
• I have read guidelines for pull request.
• My code follows the coding guidelines.
• I have performed a self code review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have made corresponding changes to the documentation (Added README.md).
• My changes generate no new warnings.
• I have added tests that prove my fix is effective or that my feature works.