e2e: fix docker network pool exhaustion

[snormore]

Summary

Fixes Docker network address pool exhaustion errors in CI:

failed to create default network: failed to create network: create network: Error response from daemon: all predefined address pools have been fully subnetted

Problem

The e2e tests create two Docker networks per test:

1. Default network - for general container communication
2. CYOA network - for client/device traffic

The default network was not specifying an IPAM config, causing Docker to allocate from its built-in address pool. Docker's default pool (172.17.0.0/12) was allocating /16 subnets, which only allows ~15 networks total. With parallel=12 tests and accumulated orphaned networks from previous runs, this pool was being exhausted.

Evidence from the CI runner:

172.17.0.0/16
172.18.0.0/16
...
172.28.0.0/16   # 10 /16 networks = nearly exhausted

Solution

1. Use separate IP ranges for each network type:

   • Default network: 10.0.0.0/8 range with /24 subnets (65,536 available)
   • CYOA network: 9.128.0.0/9 range with /24 subnets (32,768 available)

2. CI cleanup: Add pre-test step to prune orphaned networks from previous runs

The separation also fixes a test failure in TestE2E_DeviceMaxusersRollover where tests detecting interfaces by IP range (grep 'inet 9.') would incorrectly match both networks when they shared the same range.

Before vs After

Network	Before	After
Default	Docker pool (172.x/16) ~15 max	10.x.x.x/24 ~65k max
CYOA	9.x.x.x/24 ~32k max	9.x.x.x/24 ~32k max (unchanged)

Testing Verification

• go build ./e2e/... compiles successfully
• CI e2e tests pass without network exhaustion errors
• Tests that grep for inet 9. correctly identify only CYOA interfaces