Support MCP observability for Envoy AI Gateway

[wu-sheng]

Support MCP (Model Context Protocol) observability for Envoy AI Gateway

• If this is non-trivial feature, paste the links/URLs to the design doc.
  • Extends SWIP-10 per review comment requesting MCP metrics support.
  • MCP metrics defined in envoyproxy/ai-gateway mcp_metrics.go.
• Update the documentation to include this new feature.
• Tests(including UT, IT, E2E) are added to verify the new feature.
• If it's UI related, attach the screenshots below.

Changes

MAL rules (new files):

• gateway-mcp-service.yaml — 13 MCP service-level metrics (request CPM/latency/percentile, method CPM, error CPM, initialization latency, capabilities, per-backend breakdown)
• gateway-mcp-instance.yaml — 13 MCP instance-level metrics

LAL rules (modified envoy-ai-gateway.yaml):

• Split into two rules: envoy-ai-gateway-llm-access-log and envoy-ai-gateway-mcp-access-log
• LLM logs: persist error responses (>= 400) and upstream failures only
• MCP logs: persist error responses (>= 400) only
• Both rules tag ai_route_type (llm or mcp) for searchable filtering

Dashboard (modified service + instance JSON):

• Added MCP tab with 9 widgets (service) / 6 widgets (instance): request CPM, latency avg/percentile, error CPM, method CPM, initialization latency, backend breakdown

E2E test (modified):

• Added mcp-server service (tzolov/mcp-everything-server:v3 — MCP reference server with StreamableHttp)
• Added MCP request steps (initialize + tools/list + tools/call)
• Added MCP metric verification cases
• Log query uses ai_route_type=llm tag filter

Config:

• Added ai_route_type to searchableLogsTags in application.yml

• Fixed aigw healthcheck binary path (/app instead of aigw)

• If this pull request closes/resolves/fixes an existing issue, replace the issue number. Closes #.

• Update the CHANGES log.