Enhance expert load balancing logging

**Is your feature request related to a problem? Please describe.**
Cross-layer statistics: Compute a scalar balance metric per layer, then log percentiles (p50, p75, p90, p95) across all layers
Within-layer expert distribution: For each layer, log statistics of how tokens are distributed across experts (min, max, std, percentiles of tokens_per_expert)

**Describe the solution you'd like**
A clear and concise description of what you want to happen.

**Describe alternatives you've considered**
A clear and concise description of any alternative solutions or features you've considered.

**Additional context**
Add any other context or screenshots about the feature request here.