fix(extensions): correct Milvus health port and add Ollama Apple Silicon support

[yasinBursali]

What

• Milvus manifest port corrected from 19530 (gRPC) to 9091 (HTTP metrics)
• Ollama manifest gpu_backends updated to include apple at service and feature level

Why

Dashboard-api constructs health check URLs as http://{host}:{port}{health} (helpers.py:166).
Milvus serves /healthz on port 9091 (HTTP metrics endpoint), not 19530 (gRPC). The compose
healthcheck already uses http://localhost:9091/healthz — the manifest port must match or
dashboard health checks silently fail against the gRPC port.

Ollama has native Apple Silicon Metal support but its manifest only declared [amd, nvidia].
The continue service already declares [amd, nvidia, apple] as precedent.

How

• Changed service.port from 19530 to 9091 in milvus manifest
• external_port_default: 19530 preserved for user-facing gRPC access
• Added apple to gpu_backends in ollama manifest (service + feature sections)

Testing

• YAML syntax validated
• Manifest field values verified programmatically
• Critique Guardian review: approved

Known Considerations

• port is returned in ServiceStatus to the dashboard UI — Milvus will display :9091
  instead of :19530. Non-issue for dev library; a health_port schema field is recommended
  before production promotion.
• Ollama compose still declares NVIDIA GPU reservation in deploy.resources — harmless on
  macOS (Docker silently ignores), but should get an Apple overlay before production promotion.

Platform Impact

• macOS: Ollama now correctly declares Apple Silicon support
• Linux (NVIDIA/AMD): No change
• Windows: No change

Merge Order

Can merge independently — touches different lines than other open PRs on this directory.

🤖 Generated with Claude Code

[Lightheartdevs]

Approve — Two small, correct fixes.

1. Milvus port: 19530 → 9091 — health endpoint (/healthz) runs on port 9091, not the gRPC data port. Correct fix.
2. Ollama gpu_backends: [amd, nvidia] → [amd, nvidia, apple] — Ollama supports Apple Silicon via Metal. Correct.

Tiny diff (+3/-3), no conflicts. Merge anytime.

[Lightheartdevs]

Approve — Two small, correct fixes.

1. Milvus port: 19530 → 9091: Health endpoint (/healthz) runs on port 9091, not the gRPC data port. Correct.
2. Ollama gpu_backends: [amd, nvidia] → [amd, nvidia, apple]: Ollama supports Apple Silicon via Metal. Correct.

+3/-3, no conflicts. LGTM.