[codex] Track token usage per provider response

[Solvely-Colin]

What changed

• tracked normalized token and cost metadata at the provider boundary, including streaming completions
• persisted per-provider response metadata into session phase JSON and added session-level usage summaries to synthesis output
• added a new quorum usage <session> command and surfaced usage summaries in quorum session
• added CLI regression coverage for the new usage command
• fixed repo linting for top-level tests/*.ts and eslint.config.js via a dedicated tsconfig.eslint.json

Why it changed

Quorum only preserved response text, so provider usage data was lost after each call. That made it impossible to answer issue #68: token accounting per provider response, total session usage, and estimated cost reporting.

User impact

• saved sessions now retain usage metadata alongside provider responses when the upstream provider reports it
• users can inspect aggregate usage and estimated cost with quorum usage last or quorum usage <session-path>
• full session views now show the summary without needing a separate manual calculation

Root cause

Provider adapters returned plain strings, which discarded the usage payload already available from pi-ai responses. Session persistence and CLI reporting had no structure for usage metadata, so there was nothing to aggregate or render later.

Validation

• npm run build
• npm test