fix: normalize OpenAI custom URLs to prevent /v1 duplication

[roomote]

Description

This PR fixes an issue where custom OpenAI-compatible API URLs that already end with /v1 were getting another /v1 appended, resulting in /v1/v1 which caused API requests to fail.

Problem

When users configure a custom OpenAI-compatible API URL that already includes /v1 (e.g., https://custom.api.com/v1), the system was appending another /v1, creating invalid URLs like https://custom.api.com/v1/v1/chat/completions.

Solution

• Added a normalizeBaseUrl() method to the OpenAiHandler class that:

  • Ensures /v1 is present for OpenAI-compatible APIs
  • Prevents duplication if /v1 is already present
  • Preserves Azure endpoints without modification
  • Handles URLs with trailing slashes correctly

• Applied the same normalization logic to the getOpenAiModels() function

Testing

• Added comprehensive test coverage for URL normalization scenarios
• All existing tests continue to pass
• Verified with multiple URL formats including:
  • URLs already ending with /v1
  • URLs without /v1
  • URLs with trailing slashes
  • Azure OpenAI endpoints
  • Azure AI Inference Service endpoints

Related Issue

Fixes #8282

Checklist

• Code follows project conventions
• Tests added and passing
• Linting passes
• Type checking passes
• No breaking changes introduced

---

Important

Fixes URL normalization in OpenAiHandler and getOpenAiModels() to prevent /v1 duplication for OpenAI-compatible APIs, with tests added.

• Behavior:
  • Adds normalizeBaseUrl() to OpenAiHandler to ensure /v1 is present for OpenAI-compatible APIs without duplication.
  • Updates getOpenAiModels() to apply the same URL normalization logic.
  • Preserves Azure endpoints without modification.
• Testing:
  • Adds tests for URL normalization in openai.spec.ts covering various URL formats, including those with trailing slashes and Azure endpoints.
  • Verifies that URLs already ending with /v1 are not modified.
  • Ensures URLs without /v1 have it added correctly.

^{This description was created by for b736800. You can customize this summary. It will automatically update as commits are pushed.}

[ellipsis-dev]

The URL normalization logic (adding '/v1') is correctly implemented, but note that similar logic is repeated in both the normalizeBaseUrl method (lines 427–459) and in getOpenAiModels (lines 487–509). Consider extracting this logic into a shared utility function to reduce duplication and ease future maintenance.

^{This comment was generated because it violated a code review rule: irule_tTqpIuNs8DV0QFGj.}

[roomote]

Self-review: compiling my own patch like a snake eating its tail—hope the tail has tests.

Findings:
P1: Query/fragment handling in URL normalization can produce invalid URLs. Use the URL API and modify pathname only; preserve search/hash. Affects normalizeBaseUrl() and getOpenAiModels().
P2: Duplicate normalization logic between the class method and getOpenAiModels(); extract a shared helper. Also build the models URL via new URL('models', base) to avoid double slashes and preserve query.
P3: Use the normalized base for provider checks (e.g., _isGrokXAI) for consistency. Add tests covering base URLs with ?query and #fragment for both handler and model listing.