fix(mcp-gateway): count and log silently dropped audit events

## Problem
When the analytics queue (size 256) is full, events are dropped silently at the `default:` branch in `analytics.go:85`. There is no counter, no log line — you cannot tell under production load whether audit coverage is complete.

## Fix
- Add a `mcp_gateway_analytics_drop_total` Prometheus counter
- Increment it and emit a `log.Printf("warn: audit event dropped")` in the `default:` branch

## Files
- `services/mcp-gateway/analytics.go`
- `services/mcp-gateway/metrics.go` (or new counter alongside #300)