Pricing table missing current-generation models — unpriced turns bury real hotspots

[willwashburn]

Problem

The pricing table is missing several current-generation models, so a large share of turns cost $0.00 and rank invisibly in findings/hotspots — while trivially cheap but priced turns dominate the rankings.

Evidence (2026-07-30 fleet spend forensics, two independent pipelines agreeing within 1.4%)

• Unpriced models observed: claude-opus-5, claude-sonnet-5, gpt-5.6-sol, gpt-5.6-luna, gpt-5.6-terra.
• 2,652 of 4,626 turns (57%) fell under these unpriced models.
• One session burned 443M tokens in an hour and scored $0.00 severity — invisible in cost-ranked findings/hotspots.
• Meanwhile $1.26 of git status calls topped the same findings list, because it was the highest-dollar entry in a table where most of the real spend wasn't priced at all.

Fix

1. Add pricing entries for claude-opus-5, claude-sonnet-5, gpt-5.6-sol, gpt-5.6-luna, gpt-5.6-terra (see also Make scheduled workflow to update models.dev price list periodically #391, the scheduled models.dev price-list sync — this issue is the immediate gap it would have caught).
2. When a model has no price entry, findings/hotspots should rank by tokens rather than silently scoring $0.00 and sorting to the bottom. A missing price must never look cheaper than a priced trivial call.

[willwashburn]

Fresh resident reproduction after burn ingest on 2026-07-30 (~24h rolling window): burn summary --since 24h --json reported 3,479 turns and 2.999B tokens, but only $337.12. The entire priced total was Fable ($337.05) plus codex-auto-review ($0.07); claude-opus-5 (1.092B tokens), claude-sonnet-5 (13.5M), gpt-5.6-sol (1.473B), and gpt-5.6-luna (152.9M) all reported $0. This makes both dollar totals and cost-ranked hotspots unusable for the fleet question while token totals remain useful. No duplicate issue filed; this confirms #503 against the fresh resident ledger.