Improve benchmark generation script

[siddharth-krishna]

• Only generate benchmarks when something in the benchmarks/<framework>/ directory changes
• Automatically upload generated benchmarks:
  • Until the Zenodo integration is fixed, perhaps upload to a Google Cloud bucket? Or GitHub Artifact? Is there a size limit or retention limit for artifacts? Can they be downloaded by a user when using run_benchmarks.py?
• Cache pypsa input data directories so that the CI job run is faster