Improve fuzz CI matrix and artifact handling

[Psychevus]

Motivation

• Ensure the fuzz CI runs all harnesses (AES, RSA, ECIES, pipeline) across supported Python versions on schedule and manual dispatch.
• Make artifact uploads reliable so the workflow succeeds even when no crashes or timeouts are produced.
• Expose simple tuning inputs for runs/timeouts and move to current-major Actions for better long-term stability.

Description

• Expand the workflow matrix to run fuzz_aes, fuzz_rsa, fuzz_ecies, and fuzz_pipeline across Python 3.11 and 3.12 with fail-fast: false so harnesses run independently.
• Add workflow_dispatch inputs runs and timeout_seconds and wire them to environment variables FUZZ_RUNS and FUZZ_TIMEOUT_SECONDS for easy tuning without editing the YAML.
• Run each harness in a per-harness artifact directory under fuzz/artifacts/<harness> with a timeout wrapper so crash/leak/timeout files are colocated and discoverable for upload.
• Upgrade Actions to actions/checkout@v4, actions/setup-python@v5, and actions/upload-artifact@v4, and use if-no-files-found: ignore for upload to avoid failures when no artifact files exist.
• Add fuzz/README.md with simple smoke commands to run one harness or all harnesses locally.

Testing

• Ran a local smoke invocation python fuzz/fuzz_aes.py -runs=1, which failed in this environment with ModuleNotFoundError: No module named 'atheris' as expected when dependencies are not installed, indicating the harness script runs but requires the Atheris dependency to execute successfully.