fix: private IPFS swarm nodes never peer, breaking shared storage

[EnriqueL8]

Summary

• Private-mode IPFS nodes were pinned to ipfs/go-ipfs:v0.10.0, which has a private-network (pnet) connection bug where swarm connections drop immediately after handshake - nodes never actually peer with each other. This surfaces as failed to bootstrap (no peers found) and shared storage downloads timing out fetching content pinned only on another member's node.
• Bumps the IPFS image to ipfs/kubo:v0.42.0. Modern Kubo needs its AutoConf/public-bootstrap defaults disabled to even start in private-network mode, so a container-init.d script now does that for private-mode stacks.
• mDNS auto-discovery, which worked locally, proved unreliable on native Linux Docker bridge networks - members' nodes never found each other there. Added an explicit peering step after first-time-setup that queries each member's real PeerID via its own API and calls swarm/peering/add against every other member, so nodes connect (and persistently reconnect) regardless of whether mDNS works in a given environment.
• Also fixes two related bugs found in internal/docker/docker.go's output-capture helper (used to build the AutoConf-disable init script): the loop exited as soon as either stdout or stderr closed rather than both, and the final chunk of output was dropped when it didn't end in a newline - together these could silently truncate captured command output.

Test plan

• Verified locally: fresh 2-member private-mode stack peers automatically, cross-node ipfs cat and gateway fetch both succeed
• go build ./... passes
• CI Build/E2E