fix(desktop): simplify Databricks model labels

[klopez4212]

Summary

• Render Databricks gateway model IDs as clean model names in the desktop UI.
• Keep the original endpoint IDs unchanged for runtime selection.

Testing

• pnpm check
• pnpm test
• pnpm exec tsc --noEmit

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ed6043f26c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Chessing234]

Clean display-only cleanup for Databricks gateway IDs — keeping the raw endpoint ID for selection while showing a human model name is the right split. Tests around agentCardModelLabel / formatAgentModelLabel pin it. LGTM.

[Chessing234]

Approving the Databricks model label display cleanup.

[wesbillman]

Disclosure: Carl, an AI agent, is posting this comment on Wes's behalf.

I found several presentation gaps that should be addressed before merge:

1. The main Settings → Agents model field can still show raw Databricks IDs. AgentConfigFields.tsx constructs Default model (${fallbackModel}) without formatModelDisplayName and enables keepSelectedModelValueLabel. In agentConfigControls.tsx, that makes the closed picker prefer raw trimmedModel over the matching discovered option's formatted label. A configured or inherited databricks-gpt-5-5 therefore still displays with its gateway prefix.

2. Two other persisted-model surfaces bypass the formatter:

   • AgentAiConfigurationMode.tsx (HarnessModelDefaultNotice)
   • AgentAiDefaults.tsx (AI-defaults summary)

   These conflict with the new AGENTS.md rule that any user-facing Databricks model name goes through the shared formatter.

3. The heuristic mangles real endpoint families:

   • databricks-qwen3-embedding-0-6b → Qwen3 Embedding 0 6B instead of Qwen3 Embedding 0.6B
   • databricks-openai-gpt-5 → OpenAI GPT 5 instead of OpenAI GPT-5, because the GPT fix-up is anchored to the beginning of the full label

Please route the missed surfaces through the formatter, derive the stable selected label from the matching formatted option, fix the two real-ID cases, and add regression coverage. I verified that selection and persistence still use raw model IDs; this finding is presentation-only, but presentation consistency is the purpose of this PR.