Add Novita AI provider

[edelauna]

Add Novita AI provider

Problem

Novita AI (an OpenAI-compatible LLM provider, https://novita.ai/model-api/product/llm-api) is not yet available as a supported provider. Users who want to use Novita's models (e.g. moonshotai/kimi-k2.7-code) currently have no way to configure it in the extension, CLI, or webview settings.

Context

This reimplements the provider-add portion of PR #697 (closed), which bundled this work with an unrelated shared AI-SDK streaming rework. This PR should be rebased on top of the shared-streaming-fix PR (sub-issue: AI-SDK tool-call streaming/error-handling gaps) so NovitaHandler inherits the fixed base-class behavior rather than reintroducing the same gaps.

Desired behavior

• Novita AI is registered as a provider across: packages/types (model registry, provider schema, secret-key list), src/api/providers (API handler extending OpenAICompatibleHandler), src/shared/ProfileValidator, the CLI (apps/cli: --provider novita, NOVITA_API_KEY env var), the webview settings UI (base URL + API key fields, masked as password), and translated locale strings.
• e2e coverage exercises a real tool-using task through the Novita provider, including error/failure-mode paths (not just the happy path).
• Documentation (apps/cli/README.md, apps/vscode-e2e/AGENTS.md) reflects Novita as an explicit alternative to the default provider, not as if it were the default.

Constraints / preferences

• Do not reintroduce the issues fixed upstream: missing supportsTemperature on model definitions, completePrompt ignoring defaultTemperature, or an e2e completion assertion that's conditionally skipped rather than asserting the exact expected value.
• novitaBaseUrl should have at least .url() format validation (consistent with what was previously added), documented as an accepted-risk pattern shared with other free-form-endpoint providers (openai/deepseek/poe) rather than a stricter closed-enum (which only applies to fixed-region single-vendor APIs like Moonshot/MiniMax).
• Avoid unused exports (e.g., a NovitaModelId/*DefaultModelInfo pattern is fine if it matches sibling-provider convention, but shouldn't add new zero-consumer exports beyond that convention).
• AGENTS.md should explain why the e2e mock base URL suffix (/v1) differs from the real API suffix (/openai), not just state that they differ.

Acceptance criteria

Given a user configures novitaApiKey and selects a Novita model
When they send a message
Then the request is sent to Novita's OpenAI-compatible endpoint and a response streams back correctly, including tool-call handling

Given the e2e suite runs in mock mode
When the Novita test suite executes
Then it asserts the exact expected completion text unconditionally (or an explicitly-justified, narrowly-scoped exception), not a conditionally-skipped check

Request checklist

• I've searched existing Issues and Discussions for duplicates
• This describes a specific problem with clear context and impact

Related

• Tracking issue: [Tracking] Add Novita AI provider support #796
• Depends on: Fix shared AI-SDK tool-call streaming/error-handling gaps #797 (should merge first)
• Originally implemented in PR Add Novita AI provider support #697 (closed)