feat(providers): add Nscale and Vultr presets

[olddonkey]

Summary

• add first-class nscale and vultr API-key presets using their official OpenAI-compatible Chat Completions endpoints
• discover models through authenticated /v1/models requests with 256 KiB / 256 raw-row ceilings and exact official tool-model allowlists
• preserve same-named custom-provider destinations, retain provider-native model ids and flattened aliases, and keep unverified reasoning/parallel-tool claims disabled
• add fixture-only routing, discovery, validation, metadata, derivation, and collision tests; sync provider counts and scope notes across all five documentation locales

Provider evidence and scope

Provider	Official transport evidence	Agent-capable model policy	Contracting entity / routing evidence
Nscale	Chat guide, models API, and chat completion API establish the first-party Bearer-authenticated endpoint.	Nscale's catalog mixes chat, image, and embedding rows without a modality discriminator. Discovery admits only meta-llama/Llama-3.1-8B-Instruct, the exact model in the official tool-calling API example.	Nscale's terms identify Nscale AS and expressly describe its on-demand public-cloud inference offering as using third-party models/products. Its AUP applies to customers and their respective end users, supporting customer-integrated routing subject to the AUP and incorporated model licenses.
Vultr	Serverless Inference chat guide, tool-calling guide, and key retrieval establish the first-party Bearer-authenticated endpoint.	Vultr states that tool calling is currently supported only by kimi-k2-instruct, so other catalog rows remain hidden.	Vultr's pipeline-integration guidance explicitly tells customers to replace an existing inference base URL with Vultr's and use the subscription API key; the tool-calling guide covers applications and external API integrations. The service-wide Constant terms govern customer and end-user service use, while official billing instructions identify The Constant Company, LLC.

Both presets use Authorization: Bearer <key>, reject redirects during key validation, keep parallelToolCalls: false, and advertise no reasoning ladder until provider-specific tiers are documented. Maintenance owner: @olddonkey; no affiliation with either provider.

Verification date: 2026-08-03.

Sequencing

This PR remains Draft behind #870 and #872. It is currently based on standalone dev totals of 71 presets / 60 key presets; after those two batches land, a final restack will regenerate the sequential totals as 75 / 64.

Security boundary

This PR adds two canonical credential destinations, so explicit maintainer security review is requested. Vultr's publicly linked legal terms are service-wide rather than a Serverless-Inference-specific addendum, so the contracting/routing evidence is stated explicitly above for maintainer judgment. Registry-only discovery and collision policy is not persisted into config.json. Existing same-named custom providers retain their configured host and adapter, so a stored key cannot be silently redirected to either new endpoint.

Validation

• bun run typecheck
• bun test --isolate tests/nscale-vultr-provider.test.ts tests/provider-registry-parity.test.ts tests/provider-model-discovery-contract.test.ts — 58 pass, 0 fail
• bun test --isolate tests — 7597 pass, 8 skip, 0 fail across 505 files
• bun run privacy:scan
• bun run build:gui
• cd docs-site && bun run build
• git diff --check

Relates to #572. This PR intentionally does not close the umbrella issue.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: d6e0d6b1-2f7e-4454-b61f-80c76c2a7646

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[olddonkey]

Author update — current head d717f77e

The canonical-preset evidence requested by the umbrella workflow is now explicit in the PR description:

• Verification date is recorded.
• Nscale's contracting entity, third-party-model terms, and customer/end-user AUP are linked.
• Vultr's official existing-pipeline and tool-calling guidance is linked as routing authorization, together with The Constant Company, LLC evidence.
• The description explicitly notes that Vultr's publicly linked terms are service-wide rather than a Serverless-Inference-specific addendum, leaving that judgment visible for maintainer security review.
• There are currently no review threads.

The previously reported validation remains green: 58 focused tests, full suite 7597 pass / 8 skip / 0 fail, typecheck, privacy scan, GUI build, docs build, and git diff --check.

This PR remains Draft behind #870 and #872. After those batches land, I will perform the final restack and regenerate the sequential 75 / 64 provider totals. It relates to #572 and does not close the umbrella issue.