fix(ai): tolerate trailing /v1 in custom model base URL to avoid /v1/v1 requests

[solitude-mix]

Fixes #1888

What changed

• AiModelConfigServiceImpl.resolveRuntimeModel (the single place AiRuntimeModel is built, covering saved configs, request overrides, and the OPENAI_BASE_URL/ANTHROPIC_BASE_URL env fallbacks) now strips a trailing /v1 (and trailing slashes) from the base URL for the OPENAI and CLAUDE providers.
• The connection test (testOpenAiCompatibleConfig) now builds the exact URL the runtime requests (normalized base + /v1/chat/completions), so a passing test no longer masks a failing chat request.
• The ai upstream target log in AiModelFactory now prints the URL actually requested (it previously logged base + /chat/completions while the real request was base + /v1/chat/completions).
• The Base URL hint in all five locales now says the trailing /v1 is optional and never duplicated.
• New AiModelConfigServiceImplBaseUrlTest covers both providers, prefixed gateways (.../compatible-mode/v1), and non-/v1 suffixes (.../v1beta stays untouched).

Why

The settings UI instructs users to include /v1 in the Base URL, and "Test connection" honors that contract (baseUrl + "/chat/completions"). But the actual chat request goes through Spring AI's OpenAiApi, whose default completionsPath is /v1/chat/completions (Spring AI 1.1.4). Following the UI hint therefore produces .../v1/v1/chat/completions → 404 Not Found, while the connection test passes. AnthropicApi has the same trap (/v1/messages).

Stripping the trailing /v1 — rather than setting completionsPath("/chat/completions") — was chosen deliberately: it fixes configs that follow the UI hint without breaking configs that already work today with a base URL that omits /v1, and it keeps prefixed endpoints like https://dashscope.aliyuncs.com/compatible-mode/v1 working.

How I tested it

• mvn -pl :chat2db-community-domain-core -am -Dtest='AiModelConfigServiceImpl*' -Dmaven.test.skip=false -DskipTests=false -Dsurefire.failIfNoSpecifiedTests=false -Dmaven.test.failure.ignore=false test → 19 tests, 0 failures (new BaseUrlTest 7 + existing StorageTest 12).
• mvn -pl :chat2db-community-web -am ... test → compiles clean (main + test sources).
• Manually reproduced before the fix: Base URL with /v1 → 404 at .../v1/v1/chat/completions; without /v1 → works.
• Frontend change is locale-string-only; yarn lint / build:web:community were not run locally.

Known limitations

• An OpenAI-compatible endpoint whose path contains no /v1 at all (e.g. a gateway exposing /api/chat/completions) still gets /v1/chat/completions appended — same behavior as before this change, so no regression.
• The Pro edition shows the same 404 and is not covered by this community fix (see Custom OpenAI-compatible model requests .../v1/v1/chat/completions (404) when Base URL includes /v1 as instructed #1888).