Fix empty retrieval and API error handling in chat responses

[tisha-varma]

Fixes #15

This PR improves chat robustness by:

• Returning a clear message when no retrieval results are found
• Handling LLM / API failures with a user-friendly fallback
• Avoiding exposure of internal error details

The changes are intentionally minimal and scoped to Issue #15.

[Areeba-Tahir-18]

Hello !! Have you paid or add any billing card for using vertx ai while setting this project to local @tisha-varma

[tisha-varma]

Hello !! Have you paid or add any billing card for using vertx ai while setting this project to local @tisha-varma

No billing needed. This fix just adds safety checks (empty result handling + try/catch for errors). I didn't need to run any actual API calls.

[Areeba-Tahir-18]

Have you configured Vertex ai while setting project on local pc ?@tisha-varma

[Areeba-Tahir-18]

And tell me one thing more the issue you fixed is assigned by someone to you ? or you contributed by own ? @tisha-varma

[tisha-varma]

And tell me one thing more the issue you fixed is assigned by someone to you ? or you contributed by own ? @tisha-varma

I contributed on my own.

[Areeba-Tahir-18]

Are you contributing here for GSOC 2026 ?

[QuantumByte-01]

Good fix — adds an empty-result guard, wraps the Gemini call in try/except, and removes internal error details from user-facing responses. Addresses issue #15 cleanly without over-engineering. Also bundles the missing 'cd backend' README instruction which is a recurring pain point. Merging.