fix: below+bidirect edit distance takes max across strands

[aviezerl]

Summary

• Fixed direction="below" with bidirect=TRUE taking min across strands instead of max. A genomic substitution changes both strands simultaneously, so disrupting a motif requires both strands to fall below the threshold — the harder strand determines the answer.
• Rewrote edit distance documentation into a dedicated section with clear direction/strand semantics, parameter reference, and worked examples.

Details

In evaluate_windows(), the maybe_update_min lambda always kept the global minimum edit distance across all (position, strand) pairs. For direction="above" this is correct (a match on either strand suffices). For direction="below", the correct per-position combination is max across strands, then min across positions.

New below_bidirect code path computes both strands at each offset, combines with max (NaN-aware: if only one strand passes score.min, uses that strand alone), then feeds the combined result into the existing global-min tracking.

Test plan

• Existing test updated: bidirectional assertion changed from min(fwd, rev) to >= max(fwd, rev)
• New test: 1bp iterator verifying bidi == max(fwd, rev) at each position
• All 1280 PWM tests pass
• Validated on PrimatesAnc069 CTCF query — hits correctly show bidirectional scores above score.min