fix: charge new text models at a published standard tariff

[halfaipg]

Unknown text models now use USD 0.075 input / USD 0.30 output per million tokens. Explicit model prices and aliases take precedence; known media-only/zero-rate entries and unresolved auto remain ineligible. Media still requires explicit recipe/modality pricing. Quotes and durable reservation snapshots share the effective resolver; in-flight holds retain their original rate. Public pricing, OpenAI model discovery, and status expose effective rates/source. No reward registry, multiplier, payout policy, or validator authority changes. Tests cover default and explicit pricing, unknown-model credit rejection, reservation/settlement/refund idempotency, post-reserve overrides, service ceilings, discovery, auto admission and quote consistency. Local full Grid suite: 1670 passed, 495 environment-dependent skips; required PG16 CI must pass before deployment. DOX and billing audit contract updated. No migration.

[halfaipg]

DEPLOYED: immutable main e374122, Core active 2026-09-17 22:25:11 UTC. Exact-main CI35280615562 passed (2154 passed / 12 skipped), fresh production restore/parity passed, generation drained before restart, all seven workers recovered. Six processes match config. Charging and daily free spending remain on; payout controls, campaign contracts and zero-active-observation posture unchanged. Read-only billing reconciliation is clean. Public price book/default tariff and both model lists verified at 22:25:28 UTC; prior unauthenticated 401/422 responses preserved. Runtime unlisted-name quote is 375 micro-USD for 1000 input + 1000 output; unknown media remains unpriced. No customer spend or model/backend changes. Chat price tooltip is a separate undeployed PR.