Modify connected component filtering code to export summary statistics for additional transcript-level QTLs

[kauralasoo]

Introducton: For transcript-level QTLs (tx, txrevise, exon and leafcutter) we filter univariate summary statistics based on fine mapped credible sets. For each independent fine mapped credible set (cs size below threshold, not overlapping any other cs for the same gene) we currently export univariate summary statistics and lbf_variable matrices for the most strongly associated molecular_trait (i.e. leafcutter intron).

Problem: Sometimes there is a strong transcript-level QTL signal but SuSiE does not identify any credible sets (e.g. because of too much LD between variants no credible sets survives purity filtering). One such examples is the exon-level QTL for HMGCR that we highlighted in the eQTL Catalogue paper.

Proposed solution: If the gene has a significant transcript-level QTL (p-value below our FDR threshold) but now credible set is detected (or all credible sets are bigger then our size threshold for connected component analysis), then take the lead variant and most strongly associated molecular trait form the permutation analysis (used for fine mapping) and export univariate summary statistics and lbf_variable values for that moelcular trait.