Support custom base URL for OpenAI Responses API

[chyax98]

Closes #343

Problem

The native openai-api provider uses the OpenAI Responses API, but its base URL was hardcoded to https://api.openai.com/v1. Users with a local or proxied Responses API endpoint could only configure openai-compatible, which is the wrong protocol path because it targets OpenAI-compatible chat/completions rather than native /responses.

Summary

• Allow native openai-api Responses API requests to use a custom API base URL.
• Read the custom base URL from JCODE_OPENAI_API_BASE, OPENAI_API_BASE, or OPENAI_BASE_URL, including values saved in openai.env.
• Support non-interactive setup via:

jcode login openai-api \
  --api-base http://127.0.0.1:8317/v1 \
  --api-key '<key>' \
  --model gpt-5.5

• Keep openai-compatible separate and unchanged.
• Keep ChatGPT/OAuth mode on the existing ChatGPT backend URL.

Edge cases and tradeoffs

• The custom base URL only affects non-ChatGPT native OpenAI API-key mode. ChatGPT OAuth/subscription mode still uses https://chatgpt.com/backend-api/codex.
• URL validation reuses the existing normalize_api_base policy, so HTTPS is accepted and private/local HTTP endpoints are allowed, while unsafe public HTTP endpoints are rejected.
• Env precedence is JCODE_OPENAI_API_BASE, then OPENAI_API_BASE, then OPENAI_BASE_URL.

Validation

• cargo check -p jcode-base -p jcode --bin jcode
• Built local debug binary and verified native openai-api against a local Responses endpoint:

/Users/xd/.jcode/source/jcode/target/debug/jcode \
  --socket /tmp/jcode-openai-base-test.sock \
  -p openai-api -m gpt-5.5 \
  run 'Reply exactly JCODE_RESPONSES_BASE_OK'

Result:

JCODE_RESPONSES_BASE_OK