Update ilisi and add functional marker based metrics

[ghar1821]

Describe your changes

Summary

New metric metrics/functional_marker_preservation: evaluates whether batch integration preserves biologically meaningful differences in functional marker expression between two groups (e.g. WT vs KO):

• functional_marker_preservation_wilcoxon: proportion of (marker, cell type) pairs which differences are statistically significant (through Wilcoxon rank-sum) between groups unintegrated that remain significant in post integration.
• functional_marker_preservation_cohens_d: mean absolute change in Cohen's d effect size, computed for every pair that is significant (from wilcoxon above) in the unintegrated baseline.

Updated metrics/lisi: iLISI is still computed globally across all cells per split (not per group), but is now skipped and returns NaN for a split if batch is confounded by group in any group — i.e. a group whose cells all come from a single batch, so cross-batch mixing can't be assessed for it. The iLISI/cLISI computation was moved out of script.py into a new helper.py (compute_ilisi, compute_clisi, _check_batch_group_confounding), and the output uns was expanded to include per-cell iLISI/cLISI values and cell IDs for each split.

Updated methods/harmonypy: switched to the PyTorch GPU backend (harmonypy≥0.2.0) using new openproblems/base_pytorch_nvidia:1.0.0 image, automatic GPU detection, removed the manual epsilon workaround, and updated the Nextflow label to lowcpu, gpu.

Updated metrics/flowsom_mapping_similarity: mapping similarity is now computed bidirectionally (split1→split2 and split2→split1), and per-donor cluster×cell_type absolute-difference matrices are exported in the output uns.

Added scripts/adhoc_runs/run_metric_adhoc.py + .sh which re-runs a single metric against existing pipeline output by injecting a VIASH block and running it as a subprocess, discovering (dataset, method) pairs automatically, with a skip-existing option. Lets you iterate on metric code without re-running the full Nextflow pipeline.

Added scripts/fetch_intermediate_files.py which consolidate the previous two-step log-parsing/file-copying process into one script that auto-detects SLURM vs AWS Batch runs and organizes output under <dataset>/method_out/ and <dataset>/metric_out/.
This replaces the old find_intermediate_files.py and download_intermediate_files.py and copy_intermediate_files.py.

Checklist before requesting a review

• I have performed a self-review of my code

• Check the correct box. Does this PR contain:

  • Breaking changes
  • New functionality
  • Major changes
  • Minor changes
  • Bug fixes

• Proposed changes are described in the CHANGELOG.md

• CI Tests succeed and look good!