Refactor the baseline Validation Pipeline to align with GSoC Architecture

[ayushman1210]

This issue tracks the follow-up changes required for the dataframe-first validation framework introduced in PR #3892 aligning it with the architecture outlined in my GSoC 2026 workplan

Planned Changes

1. Modular 4-Stage Pipeline: Refactor the monolithic run_benchmark() entry point into a flexible 4-stage pipeline: Validate -> Align -> Compute -> Plot.
2. data.table Integration: Replace the existing naive alignment logic with high-performance data.table operations for spatial and temporal alignment, adopting patterns from PR data prep and post bias correction for NEE and LE #3934.
3. BETYdb Decoupling: Ensure that the pipeline is completely independent of BETYdb IDs, relying solely on standard variable names and tabular data structures.
4. Generalized Metrics: Expand the metric computation (RMSE, MAE, R2, correlation) to ensure they operate cleanly on the aligned data frames, independent of legacy database structures.

Context

PR #3892 established the baseline dataframe-first concept. The changes tracked in this issue will evolve it into the generalized, high-performance toolkit planned for Phase 2 of my GSoC project.

[infotroph]

data.table Integration: Replace the existing naive alignment logic with high-performance data.table operations

Please convert this to a dplyr or base R approach. data.table is a good package that lots of people use happily, but it is not widely used in PEcAn and its syntax is confusing to folks who don't know it already.