[Gap]: Still no MCP health checks despite MCP being enabled

The health response includes `"mcp": true` in the `configuration` section, but there are no MCP-specific health checks. Only REST and GraphQL entity checks appear.

## Expected

If MCP is enabled and entities are exposed via MCP, MCP endpoint health checks should appear in the `checks` array with tags like `["mcp", "list_tools"]`, similar to how REST and GraphQL each get their own checks.

| Check | Tags | What it validates | Notes |
|-------|------|-------------------|-------|
| **list_tools** | `["mcp", "list_tools"]` | Server up, tools registered, expected count based on what is enabled in the config | Baseline. If this fails, skip the rest. |
| **describe_entities** | `["mcp", "describe_entities"]` | Schema/metadata generation, pagination based on what is enabled in the config | Catches schema generation bugs. Entity count in `data` field is a good signal. |
| **read_entity** (per entity) | `["mcp", "read_entity"]` | Full data retrieval path, top 1 | Mirrors REST/GraphQL per-entity checks. Validates JSON-RPC param binding + result serialization. |
| **aggregate_entity** (per entity) | `["mcp", "aggregate_entity"]` | Aggregation code path, COUNT(*) | Genuinely different query generation than read. Cheap query, high signal. |

** MCP checks would ideally reuse one SSE connection across all MCP checks within a single health evaluation. That keeps the overhead comparable to the REST/GraphQL checks combined rather than multiplied by connection setup cost.

## Actual

The `checks` array only contains:
- Data source checks (`tags: ["data-source"]`)
- REST entity checks (`tags: ["rest", "endpoint"]`)
- GraphQL entity checks (`tags: ["graphql", "endpoint"]`)

No MCP checks exist. Combined with the missing MCP metrics from the OTEL report, MCP remains the least observable API surface.

Locate the health response and the existing REST and GraphQL entity-check generation, then trace how enabled entities are read from configuration. Add MCP checks for list_tools, describe_entities, read_entity, and aggregate_entity, including the stated tags and baseline behavior, while considering reuse of one SSE connection. Done means the checks appear only when MCP entities are enabled and validate the listed paths.

Written by the indexing model from the issue text.