Skip invalid GWAS in focal mode; validate joint GWAS inputs

[kal26]

Summary

• Joint mode: filter to valid GWAS studies; skip joint analysis if none are valid.
• Separate (focal) mode: skip invalid GWAS studies and continue; catch and log errors per study.

Changes

• Add validation for each GWAS: require z, n, variant, finite n, non-all-NA z, LD matrix present/finite, and overlapping variants with LD dimnames.
• Joint mode now runs only on validated studies; if all invalid, emit message and skip.
• Separate mode wraps colocboost(...) in try/catch and skips invalid studies with informative messages.

Rationale

Prevents single malformed GWAS inputs from halting the pipeline, which is common in heterogeneous public datasets.

[danielnachun]

Unit tests are failing here so I'm converting this to a draft so we don't merge until the problems are fixed.