Add retry-with-backoff for RequestServiceIdentity at startup #126

User Request

Gateway crashes (CrashLoopBackOff, exit code 1) when ziti-management isn't ready at startup. This is a latent startup race condition. Add retry logic matching llm-proxy's existing pattern.

Specification

Root Cause

cmd/gateway/main.go calls zitiMgmtClient.RequestServiceIdentity() at startup and uses log.Fatalf on failure. If ziti-management's gRPC server isn't accepting connections yet, the gateway exits immediately and enters CrashLoopBackOff.

Change Required

File: cmd/gateway/main.go

Wrap the RequestServiceIdentity call in a retry loop with exponential backoff:

• Initial backoff: 1s
• Max backoff: 15s
• Multiplier: 2×
• Total timeout: Controlled by ZITI_ENROLLMENT_TIMEOUT env var (default 2m)
• Retryable codes: Unavailable, Unknown (connection refused, DNS failure)
• Non-retryable: All other gRPC codes — fail immediately

Reference agynio/llm-proxy cmd/llm-proxy/main.go which already has retryWithBackoff() implementing this exact pattern (though with 5m default — use 2m).

Add log messages for each retry attempt so startup progress is visible.

What Does NOT Change

• internal/zitimgmtclient/client.go — remains a thin pass-through
• No new dependencies needed