Shard CI tests across 4 parallel jobs per backend to reduce wall-clock time

[laxmareddyp]

Description of the change

Following up on #2856 which brought CI test time from 6+ hours to ~20 min by switching to self-hosted runners and adding pytest-xdist parallelization, this PR adds test sharding to further reduce wall-clock time.

Changes

.github/workflows/actions.yml

Add shard: [0, 1, 2, 3] to the test matrix — each backend (TF, JAX, Torch) now spawns 4 parallel jobs, each running ~25% of the test suite using pytest-shard
Each shard still uses -n auto (pytest-xdist) internally for core-level parallelism within the job
Integration tests run only on shard 0 to avoid duplicate execution
Special version jobs (keras-3.15, keras-nightly, openvino) run as single un-sharded jobs (already fast enough)
requirements-common.txt

Add pytest-shard dependency for --shard-id / --num-shards support

Reference

Colab Notebook

Checklist

• I have added all the necessary unit tests for my change.
• I have verified that my change does not break existing code and works with all backends (TensorFlow, JAX, and PyTorch).
• My PR is based on the latest changes of the main branch (if unsure, rebase the code).
• I have followed the Keras Hub Model contribution guidelines in making these changes.
• I have followed the Keras Hub API design guidelines in making these changes.
• I have signed the Contributor License Agreement.

[gemini-code-assist]

Code Review

This pull request adds the pytest-shard package to the common requirements. The review feedback correctly points out that pytest-shard is unmaintained and has compatibility issues with newer versions of pytest, suggesting pytest-split as a more modern and actively maintained alternative.