feat(meta.analysis): in-memory trait_data/prior_distns for run.meta.analysis.pft (Modularity Part 2)

[omkarrr2533]

Follows #3978. Cuts the second disk hop in the chain — right now run.meta.analysis.pft() reloads trait.data.Rdata/prior.distns.Rdata that get.trait.data.pft() just wrote seconds earlier.

• Added optional trait_data + prior_distns. Pass both and the load() from pft$outdir is skipped. Pass only one and it errors instead of doing a half-disk/half-memory load.
• Added return_data (default FALSE). On TRUE the MA outputs get attached to the returned pft; on FALSE you get the bare pft like before. Did it this way because of your feat(db): extract get_trait_data_pft() as standalone file-I/O-free function #3978 point — attaching these to a pft that goes back into settings breaks serialization, so default stays legacy-safe.
• runModule.run.meta.analysis() left on default FALSE since that's the one writing into settings.
• Provenance saves / dbfile.insert() unchanged.
• Also fixed a broken roxygen title on run.meta.analysis.pft.

Format-agnostic on purpose — doesn't touch file formats, that's the separate saveRDS/CSV track. Independent of #3810.
Tests green: FAIL 0 | WARN 0 | SKIP 1 | PASS 116.
GSoC 2026 Week 2.