Fix slow serverless wake with stable container MACs

[arjunkomath]

Summary

• Derive a deterministic, locally administered MAC address from each static container IPv4 address.
• Pass that MAC to Podman with --mac-address when creating static-IP containers.
• Add regression coverage for MAC derivation and Podman run arguments.

Root cause

Podman generated a different MAC whenever a stopped serverless container rejoined the bridge. The proxy retained the old IP-to-MAC neighbor entry, so Linux waited for neighbor discovery to fail before resolving the new MAC. This produced the repeatable attempts=12, roughly 8.9-second wake latency.

Impact

New and redeployed static-IP containers retain the same MAC across sleep/wake cycles, removing the stale-neighbor delay. Existing containers need one controlled redeploy to receive the creation-time MAC. Health-check and readiness behavior are unchanged.

Validation

• go test ./...
• OrbStack integration test with Podman 5.4.2 across two sleep/wake cycles:
  • wake latency dropped from roughly 8.9 seconds to 626ms and 1.166s
  • container ID and MAC stayed unchanged
  • neighbor state transitioned normally without FAILED
  • both direct-IP and loopback upstream probes succeeded