check_guest_connections fires TEST_CONNECTION_COUNT (50) concurrent
guest->host connects, but host_echo_server listened with backlog=5.
The vsock muxer establishes the host-side connection with a blocking
connect() on the VMM event-loop thread; when the backlog saturates
(the whole worker burst reconnecting at once after a snapshot
restore), that connect() stalls the VMM and delays OP_RESPONSEs past
the guest's 2s vsock connect timeout, causing intermittent
"connect(): timed out" failures in test_cycled_snapshot_restore.

Use the existing SERVER_ACCEPT_BACKLOG (128) so the backlog comfortably
exceeds the concurrent connection count, matching the guest-side
listener.

Signed-off-by: Riccardo Mancini <mancio@amazon.com>