Guard _mirrored_baf against all-NaN median to fix 3.11-min CI (gh#407)

[etal]

Summary

master is currently red on the 3.11-min CI job (e.g. the merge of #1077, run 26365341592), and every PR branched off master inherits the failure (#1078, #1079). This fixes the root cause.

Root cause

PR #407 keeps alt_freq as NaN when a VCF lacks allele counts. A bin whose heterozygous SNPs all lack frequencies then hands _mirrored_baf an all-NaN slice, where it calls:

above_half = vals.median() > 0.5   # cnvlib/vary.py

Under the min-pinned (older) numpy, pd.Series.median() on an all-NaN slice routes through np.nanmedian → np.nanmean of an empty slice, which emits RuntimeWarning: "Mean of empty slice". Because pyproject.toml sets filterwarnings = ["error"], that warning becomes a test failure. Newer numpy (the regular jobs) doesn't warn, so only 3.11-min fails.

#407 added an all-NaN guard in the outer summarize() (protecting summary_func), but the inner median() in _mirrored_baf runs first and was left unguarded.

Affected tests (both via baf_by_ranges → summarize → _mirrored_baf):

• test/test_commands.py::CallTests::test_call_filter
• test/test_io.py::IOTests::test_read_vcf_het_no_alt_count

Fix

Short-circuit _mirrored_baf for an all-NaN slice before calling .median(); the mirrored result is all-NaN regardless of mirroring direction:

if vals.isna().all():
    return shift   # all-NaN
above_half = vals.median() > 0.5

Tests

• New VATests::test_mirrored_baf_all_nan (test/test_cnvlib.py): under warnings.simplefilter("error"), asserts an all-NaN input mirrors to all-NaN without warning, and that a partially-NaN slice still mirrors correctly (median taken over the real values).
• Locally (numpy 2.4.3) test_cnvlib.py + test_io.py + test_commands.py = 119 passed; mypy and ruff clean. The warning is numpy-version-dependent, so the warnings-as-error assertion is the guard that bites in the 3.11-min environment.

Clinical-impact note

Behavior is unchanged for VCFs that do carry allele counts (the median() path is untouched there). For the no-allele-count case the BAF was already meant to be NaN (#407); this only stops a spurious warning/crash. No change to .cnr/.cns/.cnn/SEG/VCF numeric output.

Refs #407.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.81%. Comparing base (c479179) to head (9162f9e).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1080   +/-   ##
=======================================
  Coverage   67.80%   67.81%           
=======================================
  Files          74       74           
  Lines        7684     7686    +2     
  Branches     1365     1366    +1     
=======================================
+ Hits         5210     5212    +2     
  Misses       2034     2034           
  Partials      440      440           

Flag	Coverage Δ
unittests	67.81% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.