refactor CHIRPS3 DHIS2 workflow into modular processes

[abyot]

Summary

Refactors the CHIRPS3 -> DHIS2 flow from a monolithic processor into modular, reusable OGC processes with a workflow entrypoint, plus safer cache behavior and cleaner naming.

What changed

• Replaced legacy chirps3-dhis2-pipeline with a process-first design:
  • chirps3-dhis2-workflow (assembly/orchestration)
  • feature-fetch
  • chirps3-download (reused)
  • data-aggregate
  • dhis2-datavalue-build
• Kept orchestration concerns in chirps3_workflow.py + workflow_runtime.py (run_process_with_trace).
• Made decomposed processes self-contained.
• Added flavor support end-to-end (rnl/sat) with stage/flavor validation.
• Removed deprecated chirps3_dhis2_pipeline.py.
• Updated pygeoapi process wiring and docs/diagrams accordingly.

Caching improvements

• Added shared cache utilities (src/eo_api/utils/cache.py).
• Improved CHIRPS download cache scoping by spatial+product dimensions (bbox, stage, flavor) to avoid stale reuse.
• Added delta-friendly reuse:
  • download layer reuses existing monthly files and fetches only missing months
  • aggregation layer reuses cached (orgUnit, period) rows and computes only missing spatial/temporal deltas