feat(providers): add Opper as an aggregator provider

[Felixkw12]

feat(providers): add Opper as an aggregator provider

Summary

Adds Opper to the provider catalog as an aggregator: the EU-hosted AI gateway with 700+ models from 30+ providers behind one OpenAI-compatible API and one key (OPPER_API_KEY). Users who need EU hosting or a single key across OpenAI, Anthropic, Google, Mistral, DeepSeek, Moonshot and open-weight models can pick it from the catalog instead of configuring a custom endpoint.

The entry follows the OpenRouter / ZenMux shape exactly:

• packages/core/src/models-dev-projection.ts maps opper to the models.dev provider opper, so provider facts, model metadata and pricing come from the normal snapshot pipeline.
• packages/core/src/provider-registry.ts adds the facts guard, the tool-calling fallback list (recommended: anthropic/claude-sonnet-4-6, openai/gpt-5.5, gemini/gemini-3.5-flash, moonshot/kimi-k3) and the registry row: openai-compatible runtime adapter on https://api.opper.ai/v3/compat, api_key auth, live protocol discovery (/models with the user's key; 401 without one), overseas / aggregators, catalogOrder 40.5 right after OpenRouter, signup at https://platform.opper.ai.
• apps/desktop/src/renderer/settings/provider-display-copy.ts: en / zh-CN / zh-TW copy with the Aggregator badge.
• apps/desktop/src/renderer/settings/provider-brand-marks.tsx + assets/provider-brands/opper.svg: the Opper mark as a masked monochrome asset (same path as ZenMux), with a provenance comment; renderer-architecture.json ledger updated via --write.
• Tests: a catalog contract case for the Opper row and a snapshot pricing case (opper:anthropic/claude-sonnet-4-6).

Model ids are the ones models.dev lists for Opper today (vendor/model, which pins one upstream route). models.dev is moving Opper to bare pool ids (anomalyco/models.dev#6542: claude-sonnet-4-6, Opper picks the route per request); when that lands, the regular snapshot refresh carries it over with no change here.

Refs: none. New catalog entry, no open issue.

Verification

Run on 2026-09-17 against main @ f9d3efd (rebased after #5437 changed the metadata projection; Node 25.8, npm 11.11):

• node scripts/sync-model-metadata.mjs --check — generated modules match the committed snapshot
• npm --workspace @maka/core run build and npm --workspace @maka/runtime run build — clean
• node --test --test-concurrency=4 packages/core/dist/__tests__/provider-catalog-contract.test.js packages/core/dist/__tests__/model-metadata.test.js — 37 passed
• npm --workspace @maka/core run test:dist — 853 passed
• node --test --test-concurrency=4 packages/runtime/dist/__tests__/cost.test.js — 10 passed
• node --test --test-concurrency=4 packages/runtime/dist/__tests__/model-factory-thinking.test.js packages/runtime/dist/__tests__/provider-contract-matrix.test.js — 177 passed
• npm run lint, npm run format:check, npx knip --workspace apps/desktop, npx knip --workspace packages/ui, git diff --check — clean
• pre-commit: ASF header audit and protocol epoch guard passed

Not clean on this machine, identically on pristine main @ f9d3efd with the same npm install (Node 25.8 / npm 11.11 rather than the pinned npm 11.19): npm run build and npm run typecheck stop in @maka/ui at src/primitives/module-page.tsx(149,5) (TS2769), and npm --workspace @maka/desktop run test:dist reports 54 failures out of 2408. None of those files are touched here; before the rebase (base of 2026-09-16) the full build, typecheck and desktop suite were green with this change applied (desktop 2719 passed). I am relying on CI for those three at the current base.

Not run: a live chat completion through the desktop app with an Opper key (no key on the machine used for this branch); the catalog row, brand mark and copy are covered by the build and contract tests only.

Review focus

Snapshot. scripts/model-metadata/models-dev-api.snapshot.json gains only the Opper section (40 models, 27 priced) plus the new projectionSha256; every other provider's entries are byte-for-byte as committed. A plain --refresh currently aborts before writing anything: upstream models.dev now ships fireworks-ai/accounts/fireworks/models/minimax-m2p7 with a provider override that has no npm field, which toModelProviderOverride rejects ("fails loud and whole" by design). The Opper section was therefore projected with the repository's own toMetadata / toPricing from the models.dev response retrieved 2026-09-17 13:45 UTC (ETag W/"735eea088a5e0c3a08b2b505eb43a532", SHA-256 735eea088a5e0c3a08b2b505eb43a53259307659fc77d384c64872797d98be0b) and spliced in at the position a refresh would place it. Once upstream or the projector handles that override, a normal refresh reproduces the same Opper entries. Happy to rebase onto a full refresh instead if you prefer to land that first.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude (Anthropic) drafted the registry entry, display copy, tests, snapshot splice and this description, mirroring the existing OpenRouter / ZenMux / Moonshot Global entries; a human at Opper reviewed every line, ran the checks above, and owns the submission.

Checklist

• Tests cover the change and fail without it
• Lint, format and the affected suites pass locally (typecheck / full build: see Verification — same result on pristine main with this install)

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

Notes

Opper is already listed on models.dev (opper). Token rates are the model providers' rates with no markup. Opper charges a 3% fee when you buy credits. I work at Opper; happy to adjust anything to match project conventions.