Grade 5'/3' bias symmetrically via abs(log2) distance

[rmillikin]

The report card previously penalized only low 5'/3' bias values (3' enrichment), treating a 5' enrichment of 2.0 as GOOD. Since the metric is a ratio, a value and its reciprocal represent the same magnitude of coverage non-uniformity. Grade it on the symmetric log2 distance from the ideal 1.0 instead, so both directions are penalized equally:

1.32 fail (raw <0.40 or >2.50), >1.00 warn (<0.50 or >2.00),
0.30 ok (<0.81 or >1.23), else good (0.81-1.23).

The low-side thresholds reproduce the previous edges, so the degradation direction grades as before. A per-metric grade_fn decouples the graded/ drawn quantity from the displayed value, which still shows the raw ratio.

Claude-Session: https://claude.ai/code/session_01GAnubcGrQV293bp3d7knvF