Add times_seen_threshold slider to dashboard and set default=1 in evaluate notebooks

[jaredgalloway]

Summary

• Adds a times_seen_threshold slider (default 1, range 0–20) to experiments/dashboard.py. The slider appears in both the Param Correlation and Replicate Scatter tabs and is wired into mc.mut_param_dataset_correlation(times_seen_threshold=...) and mc.split_apply_combine_muts(..., times_seen_threshold=...).
• Sets times_seen_threshold=1 at all three call sites in experiments/scv2-spike/notebooks/evaluate.ipynb (mut_param_dataset_correlation, combine_replicate_muts, split_apply_combine_muts).
• Sets times_seen_threshold=1 at both call sites in experiments/simulation/notebooks/evaluate.ipynb (split_apply_combine_muts, mut_param_dataset_correlation).
• Sets beta0_ridge: 0.0 in experiments/scv2-spike/config/config.yaml for the production run.

Motivation

Mutations whose reference-condition times_seen == 0 in one replicate land at exactly β = 0 (zero-initialized, no data to drive the estimate), producing prominent horizontal/vertical stripes in replicate-scatter panels that artificially deflate correlation estimates. threshold=1 removes these — it is the weakest filter that excludes zero-evidence mutations. Setting threshold=0 on the slider exactly reproduces pre-change behavior.

Backward compatibility

CSV outputs (library_replicate_correlation.csv, mutations_df.csv, collection_muts.csv) will have fewer rows after re-running the pipeline — the removed rows are exactly those with min(times_seen_*) == 0. Downstream consumers (manuscript_figures.ipynb) should be re-run; changing correlation values are expected and indicate the filter is working.

Test plan

• ruff check . and black --check . pass (confirmed locally)
• Launch dashboard against a spike fit_collection.pkl; verify slider renders in both tabs, threshold=0 reproduces old behavior, threshold=1 eliminates β=0 stripes
• Re-run spike pipeline; verify CSV row counts drop and no errors

Closes #235

🤖 Generated with Claude Code