Add Gaussian ILD partner-dependence diagnostics

[Pascal-Kueng]

Summary

• Extend Gaussian identity-link partner-dependence checks to intensive longitudinal dyadic data.
• Cover stable, concurrent, own-lag, and directed cross-lag summaries using one shared pair-statistics kernel.
• Support exact scheduled-time gaps plus dyad and edge weighting.
• Add focused tests, documentation, a shipped-data benchmark, and an outer simulation study.

Validation

• Complete package suite: 1,623 expectations passed without failures, warnings, or skips.
• Exact source-tarball R CMD check --as-cran completed with only the expected development-version NOTE.
• Shipped-data benchmark: 120 dyads, 3,360 fitted rows, 199 simulations, and all 32 summaries defined.
• Compact/refactored implementation matched the preceding validated implementation to floating-point precision.
• Pilot simulation: all 120 fitted scenarios converged and completed without fit, Hessian, warning, or diagnostic errors.

Review note

This is a stacked draft PR based on PR #18. It intentionally targets residual-diagnostics and should be retargeted to main after PR #18 merges. The checks remain descriptive plug-in predictive checks; they do not report p-values or claim formal calibration.