Fix rrvgo bugs in gprofiler.r; add multi-ontology reduction, treemaps,

[ronstewart]

JSON hierarchy, and significance-colored diverging plots

Ported over the bug fixes and later feature work done on scRNAseq_downstream's copy of this script (which started from this one), adapted to this script's CLI-args + optional two-file diverging-comparison design (--file2/--color1/--label1).

Three bug fixes (independent of everything else below)

• Log-base inconsistency: the raw bar plot used natural log (-log(p.adj)) while the rrvgo score used log10 (-log10(p.adj)), so the same term's bar height looked wildly different between the two plots (~15 vs ~6.5 for the same p-value) despite the axis label already claiming log10 for both. Both now use log10 consistently.
• calculateSimMatrix()'s last internal step is m[!out, !out], which R's default matrix indexing silently collapses to a bare scalar whenever exactly one term survives orgdb/ancestor filtering -- losing the dimnames that would identify which term it was. A scalar NA means truly zero terms mapped; a non-NA scalar means exactly one term mapped. The previous check (!is.null(dim(sim_matrix)) alone) treated both cases as "zero mapped," silently dropping a real, valid single-term cluster.
• orgdb was hardcoded to "org.Dr.eg.db" regardless of --organism, so e.g. running with --organism hsapiens would still (incorrectly) use zebrafish annotations for the rrvgo step. Now looked up from --organism via deseq2/build/src/organism_orgdb_map.txt (same design as scRNAseq_downstream's data/organism_orgdb_map.txt -- add a row there and install the matching Bioconductor package for a new organism). Resolved relative to this script's own location (not getwd()), since this script has no "run from repo root" convention.

Feature parity with scRNAseq_downstream, adapted for --file2

All four of these live inside run_gprofiler_analysis(), which already runs once per gene-list file -- so file1 and file2 (when given) each get their own full set of per-file outputs independently; only the final combined diverging plots needed new logic to merge them.

• Multi-ontology support: new --ontologies flag (comma-separated, e.g. "BP" or "BP,MF,CC"; defaults to "BP" to match prior behavior). Semantic similarity is only meaningful within one GO ontology at a time, so each requested ontology is run through calculateSimMatrix/reduceSimMatrix separately and stacked with an ontology column. Terms whose ontology can't be resolved, or whose ontology isn't in --ontologies, or that calculateSimMatrix couldn't map in orgdb, are recorded (with why) in a new _reduced_excluded.csv per file instead of silently vanishing.
• Significance-colored diverging plots: the combined plots used to color bars by file1-vs-file2 identity (opt$color1 vs red), which meant they couldn't also show significance, and conflicted with wanting a continuous color scale like scRNAseq_downstream's. Since bar position (left/right of zero) already encodes file1 vs file2 -- that's the whole point of the sign flip -- fill color is now a continuous -log10(p.adj) gradient (blue->red) on both sides instead, applied to a new unsigned significance column carried alongside the signed score (used only for position). No information is lost: the y-axis label already spells out which side is R/NR.
• reduced_barplot is now faceted by ontology (one panel per BP/MF/CC), and grouped by parentTerm rather than individual term, matching scRNAseq_downstream -- a term absorbed into a more significant parent's cluster no longer gets its own bar.
• Per file, per ontology with data: reduced_treemap.pdf (rrvgo's own cluster visualization -- doesn't facet, so one PDF per ontology instead of trying to force them into one), plus _reduced_grouping.csv (one row per parent cluster, listing member terms directly) and _reduced_hierarchy.json (the same grouping as nested JSON: ontology -> parent cluster -> member terms, browsable in a JSON viewer's fold view -- easier to scan than the flat CSV or a busy treemap with many clusters).

Test data + .gitignore

Added deseq2/example_data/gprofiler/ (mirroring the existing deseq2/example_data/ convention for deseq2.r's own test inputs):

• photoreceptor_vs_other_degenes.txt: real zebrafish photoreceptor marker genes, converted from scRNAseq_downstream's data/test_gprofiler/photoreceptors_vs_other_lfc0.25.txt (Seurat avg_log2FC/p_val_adj column names -> this script's expected DESeq2-style log2FoldChange/padj).
• immune_response_degenes.txt: synthetic (clearly marked as such in its header comment) immune/inflammatory-response gene data, specifically to pair with the photoreceptor file as --genes/--file2 for a genuinely asymmetric diverging-plot test -- testing --file2 against the same file twice is trivially symmetric and doesn't exercise the position/color encoding logic above.

Both organism_orgdb_map.txt and the new example_data files were silently excluded by .gitignore's existing blanket *.txt rule (no .txt file has ever been tracked in this repo). Added targeted ! exceptions for these specific paths rather than removing the blanket rule, so future generated/log .txt output stays ignored.

Verified end-to-end: single-file runs (multi-ontology, all new per-file outputs) and two-file diverging runs, including a run against real photoreceptor vs. synthetic immune data confirming genuinely asymmetric results (different terms, different magnitudes, no forced symmetry) with correct position-for-direction/color-for-significance rendering.