docs(book): add Node Operator Guide and DisTrO explainer

[salch-cred]

What

Adds two new chapters to the psyche-book:

1. Node Operator Guide (psyche-book/src/explain/operator-guide.md) — a field manual for people operating Psyche training nodes long-term:

   • what the operator actually owns (run-manager, container, keypair, RPC, GPU stack)
   • the epoch lifecycle from the operator's seat (WaitingForMembers -> Warmup -> Rounds -> Cooldown), with per-phase failure modes and how to read epoch boundaries in logs
   • how per-epoch rewards accrue (whole-epoch completion) and what that implies for maintenance windows
   • session management (tmux + systemd unit example), startup checklist, monitoring (container, phases, GPU, disk, RPC), updating (container automatic vs run-manager manual)
   • graceful exit timing, delegation/multi-machine setups, claiming via the Treasurer, and a troubleshooting quick-reference table

2. DisTrO (psyche-book/src/explain/distro.md) — an explainer for the DisTrO optimizer, grounded in the actual implementation (shared/modeling/src/distro.rs):

   • the synchronization-bandwidth problem and why data-center-style gradient sync fails over the internet
   • the two key ideas: DCT-based compression + top-k, and momentum-residual accumulation so dropped components are recovered in later rounds
   • a step-by-step walkthrough of Distro::generate (decode peers' updates -> apply -> gradient -> residual update -> DCT encode -> top-k -> 1-bit sign quantization -> P2P broadcast), with a mermaid diagram
   • how it fits with the Coordinator rounds/witness quorum (off-chain weights, on-chain state machine)
   • config knobs table (compression_decay, compression_chunk, compression_topk, quantize_1bit) and how they interact with max_round_train_time

Also wires both into SUMMARY.md (DisTrO under "Psyche in depth", Operator Guide as its own top-level section after it).

Why

• There is currently no operator-oriented documentation: the quickstart covers first boot, the FAQ covers one-liners, but nothing covers running a node over time (epochs, maintenance, monitoring, recovery). Node operators are the run's supply side, and the on-chain epoch/reward semantics (coordinator/mining-pool/treasurer programs) make the timing of joins/exits financially meaningful — that deserves explicit documentation.
• DisTrO is the core novelty that makes Psyche work over the internet, but it is only documented in blog posts; the repo itself has no explanation of shared/modeling/src/distro.rs, its wire format, or its config knobs. The explainer links the optimizer to the run config and the general workflow chapter.

All content is grounded in the repo itself: the epoch state machine and witness/quorum terminology come from general-workflow.md/glossary.md, the reward rules from rewards.md, the delegation flow from quickstart-compute-provider.md/join-run.md, and the DisTrO mechanics from the Distro/TransformDCT/CompressDCT implementations in shared/modeling/src/distro.rs. Internal links verified to resolve against the book tree.

Happy to restructure (e.g. move the operator guide under enduser/, or change section ordering) if maintainers prefer a different placement.

[salch-cred]

Closing as a duplicate - superseded by two focused PRs: DisTrO explainer in #651 and Node Operator Guide in #652. This PR's head branch now points at the operator-guide-only commit (identical to #652), so its title no longer matches its contents. The two focused PRs merge independently and avoid SUMMARY.md conflicts.