fix(pysam): retain variant summaries with non-finite QUAL values

[ketpatil77]

A local VCF containing a nan QUAL value caused variant_summary.py to exit 2 when strict JSON serialization encountered a non-finite mean. HTSlib also accepts positive and negative infinity.

Exclude non-finite qualities from the mean and count them separately as record_counts.nonfinite_quality, while preserving all variants in other counts. Document that behavior and bump the pysam skill metadata version.

Validation: reproduced the CLI failure before the fix; all three non-finite regressions failed before the change. python -m pytest tests/pysam -q: 63 passed, 33 subtests passed. Repository meta checks: 18 passed, 4,224 subtests passed. No remote genomic data was accessed.

[rahul05ranjan]

Verified the crash path and the fix mechanics locally (my host doesn't have pysam, so I couldn't run tests/pysam itself — details below).

• Confirmed the script serializes with allow_nan=False (variant_summary.py line 340), so on main a nan QUAL falls into the else branch and poisons the mean: json.dumps({"mean_quality": nan}, allow_nan=False) raises ValueError: Out of range float values are not JSON compliant. That matches the reported exit 2, and inf/-inf would hit the same wall.
• Replicated the new branch logic (lines 246–254) with a stdlib snippet: [nan, 50.0] → nonfinite_quality=1, mean_quality=50.0, serializes clean; [inf, -inf, 50.0] → nonfinite_quality=2, mean still 50.0. The all-non-finite edge is also safe: mean_quality is already guarded by if quality_count (line 308) and becomes None, which passes strict JSON.
• One observation: since record_counts is a Counter, nonfinite_quality only appears in the report when non-finite QUALs exist — but missing_quality behaves the same way, so it's consistent with the existing shape.

Nice minimal fix, and the docs paragraph plus the metadata bump cover the behavior change. LGTM.