Add provider cost dashboard

[Siddhant-K-code]

Summary

• add agent-strace cost --breakdown provider across stored Anthropic, OpenAI, AWS Bedrock, and Gemini sessions
• infer providers from recorded model IDs and calculate spend from actual input/output token counters
• bundle an explicitly dated offline pricing snapshot with official maintainer sources and no default network calls
• support ISO/relative --since filters and exact raw CSV export columns
• make --live-pricing fail explicitly when compatible authoritative pricing APIs are unavailable instead of scraping or inventing rates
• document pricing limitations/update steps and bump agent-strace from 0.84.0 to 0.85.0

Closes #208

Verification

• PYTHONPATH=src python -m unittest tests.test_cost tests.test_provider_cost tests.test_explain tests.test_timeline tests.test_telemetry -q — 139 tests passed
• Ona test task — 1,682 Python tests and 3 VS Code extension tests passed
• python -m compileall -q src/agent_trace && git diff --check
• wheel build — agent_strace-0.85.0-py3-none-any.whl

[Siddhant-K-code]

PR review completed: checked legacy pricing compatibility, provider inference precedence (including Bedrock profiles), token de-duplication, date filtering, unknown-model behavior, CSV stability, offline guarantees, pricing-source documentation, and aggregate session counts. One local review finding (total rows vs unique sessions) was fixed before opening the PR. No blocking findings remain; merge is gated on all five CI jobs.