feat(desktop): support API modes for custom endpoints

[JackLee992]

What does this PR do?

Desktop custom endpoints currently assume an OpenAI-compatible API surface: the form cannot choose a wire protocol, saves do not mirror a protocol to the active model, and the Test action only calls /models. An Anthropic-compatible relay can therefore validate or save misleadingly while real chat falls back to /chat/completions.

This adds the Desktop counterpart to the custom-provider API mode support already available in the CLI:

• expose Auto-detect, Chat Completions, Responses / Codex, and Anthropic Messages in Custom Endpoints
• persist the selection to the v12 providers.<id>.transport field while preserving existing hand-written values for older clients
• mirror the resolved mode to model.api_mode on Save/Use, clear it when Auto is selected, and remove it when the active endpoint is deleted
• make Test protocol-aware: /models for Chat Completions, a disclosed one-token /responses request for Responses, and a disclosed one-token /v1/messages request with Anthropic headers for Anthropic Messages
• localize the new UI copy in English, Simplified Chinese, Traditional Chinese, Japanese, and Arabic

Related Issue

Follow-up to #13415 and the Desktop custom-endpoint work in #67759. No dedicated open issue found.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• apps/desktop/src/app/settings/custom-endpoints-settings.tsx: add the API Mode selector and mode-specific guidance.
• apps/desktop/src/types/hermes.ts and apps/desktop/src/i18n/*: extend the API contract and localized copy.
• hermes_cli/web_models.py: validate supported Desktop API mode values.
• hermes_cli/web_server.py: round-trip/persist transport, synchronize active model state, and probe the selected protocol.
• Python and Vitest regression tests cover save, Auto clearing, older-client preservation, activation/deletion, runtime resolution, and Anthropic request shape.

How to Test

1. Open Settings → Providers → Custom Endpoints.
2. Add an Anthropic-compatible endpoint using the API root (without a trailing /v1), choose Anthropic Messages, enter a model and key, then click Test.
3. Save with Use for new chats enabled and verify providers.<id>.transport: anthropic_messages plus model.api_mode: anthropic_messages in config.yaml.
4. Start a chat and verify the provider receives POST /v1/messages rather than /chat/completions.

Automated verification performed:

• scripts/run_tests.sh tests/hermes_cli/test_web_server.py -q — 149 passed
• npm run test:ui — 406 files / 3621 tests passed
• npm run check:lint — passed (no errors)
• npm run build — passed
• python scripts/check-windows-footguns.py hermes_cli/web_models.py hermes_cli/web_server.py — passed

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this feature
• I've run the entire Python test suite (targeted Web API suite passed: 149 tests)
• I've added tests for my changes
• I've tested on macOS 26.4 arm64

Documentation & Housekeeping

• Documentation N/A — the new field has localized inline guidance and uses an existing config capability
• cli-config.yaml.example N/A — transport already exists; this exposes it in Desktop
• CONTRIBUTING.md / AGENTS.md N/A — no workflow or architecture change
• Cross-platform impact considered; changed production Python files pass the Windows footgun checker
• Tool descriptions/schemas N/A — no agent tool changed

Screenshots / Logs

No screenshot attached; the new selector is covered by a Desktop behavior test and the production Desktop build succeeds.