Add Prometheus metric for policy decisions

[saivedant169]

Problem

AegisFlow exposes Prometheus metrics for requests, latency, tokens, and errors, but there is no dedicated metric for policy decisions. Operators running this in production cannot easily graph the ratio of allow vs review vs block over time.

Acceptance criteria

• Add a new counter: aegisflow_policy_decisions_total{decision="allow|review|block", protocol="...", tool="..."}
• Increment it inside the policy evaluation path so it covers both the MCP gateway and the direct handler
• The metric shows up under /metrics on the admin port
• A unit test verifies the counter increments correctly for each decision

Files to touch

• internal/middleware/metrics.go (or a new file in the same package for policy metrics)
• Wherever the policy engine returns a decision that gets counted (internal/toolpolicy, internal/mcpgw)

How to test locally

# Start AegisFlow, trigger one of each decision
aegisctl test-action --protocol mcp --tool github.list_repos --target foo
aegisctl test-action --protocol mcp --tool github.create_pr --target foo
aegisctl test-action --protocol mcp --tool github.delete_repo --target foo

# Check metrics
curl http://localhost:8081/metrics | grep policy_decisions_total

Notes

Cardinality matters. If the tool label becomes unbounded (for example, user-supplied names), consider dropping it or keeping it in a separate metric. A safe default is to include protocol and decision but omit tool from the labels and emit the tool name as a log line instead.