Copilot's OTel LLM-call table showed aggregate input tokens but did not surface cache reads when exporters provide them. This makes smoke-test economics harder to inspect across model families.

Summary

• Parse per-call cache-read token attributes, including common aliases and nested token detail fields, into LlmCallSummary and turn/economic aggregates.
• Expand the OTel LLM calls table with cache read, total tokens, wall time, and current/limit context columns.
• Carry OTel cache-read totals into Pier metrics and documentation while preserving native shutdown totals as authoritative when available.

Testing

• uv run ruff check --fix .
• uv run ruff format .
• uv run ruff check .
• uv run pytest -q