Check for llava-tgi-service being started rather than healthy

[ashahba]

Description

For some reason, llava-tgi-service never becomes healthy and the dependent services within VisualQnA fail to start and eventually docker compose -d up fails on CPU.
This is probably a subtle bug in the way docker compose healthcheck works.

Issues

Although may not be the exact reason that the user is running into issues here #1479 but it may partially resolve that issue

Type of change

Replace docker compose readiness condition service_started instead of service_healthy

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds new functionality)
• Breaking change (fix or feature that would break existing design and interface)
• Others (enhancement, documentation, validation, etc.)

Dependencies

N/a

Tests

I can successfully start the services and make curl calls to the microservice(s).

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues found.

Scanned Files

[Spycsh]

I cannot reproduce the issue that llava-tgi-service never becomes healthy on my Xeon server. Could you docker logs llava-tgi-service and show whether it ends with router/src/server.rs:2354: Connected, which means the service is up successfully?

1479 is about VideoQnA, rather than VisualQnA? So this PR cannot solve 1479, right?