[Chase-Rong]

Description

This PR rewrites both pages from the current implementation and fills smaller
gaps in the related pages. Documentation only, no code changes.

design/conductor/conductor-architecture-design.md (rewritten)

• Components: EventManager, ConductorService (shared by HTTP and RPC),
  ZMQClient, KVEventHandler, PrefixCacheTable, and the C++/Python/Rust
  clients.
• Data model: context key (tenant_id, model_name, lora_name, block_size), one
  hash profile per context, the 200000-entry cap per context, and the
  device / cpu_local / cpu_share / disk tiers with the medium-to-tier
  mapping for vLLM, SGLang, and Mooncake events.
• Hash profiles (vllm_v1, sglang, sglang_bigram) and how block coverage
  differs between them.
• Event flow, clear scope per publisher type, gap detection, replay over
  DEALER, stale sources, and reconnect.
• Query flow with cumulative tiers and clamping to the queried token count.
• Config keys (http_server_port, rpc_server_port, kvevent_instance),
  environment variables, the WITH_CONDUCTOR CMake build, and the current
  directory layout.

api-reference/http/conductor-indexer.md (rewritten)

• Encoding is MessagePack only, with Content-Type: application/msgpack.
  Requests are validated strictly, and errors use the
  {error, reason, field, index} format.
• /register documents the required hash_profile and the optional
  cache_group. The additionalsalt field, which does not exist, is removed.
• /query documents the actual response shape:
  {"instances": {id: {longest_matched, dp, rank_matches, device, cpu_local, cpu_share, disk}}}. The old doc showed a response keyed by tenant with
  GPU/CPU/DISK fields.
• Documents GET /global_view and GET /services. Removes /query_by_hash,
  the lora_id / block_hash aliases, and the XXH3 hashing section, none of
  which are implemented.
• Replaces the Store master publisher section. The master does not parse
  object keys into seq_hashes, and it does emit model_name /
  block_size / lora_name / dp_rank from its flags. The section now
  explains how Conductor derives block identity from object_key under each
  hash profile, and the steps to index a master stream. The
  (mooncake-store-master-publisher)= anchor is kept because the Store
  deployment guide links to it.

Module

• Transfer Engine (mooncake-transfer-engine)
• Mooncake Store (mooncake-store)
• Mooncake Conductor (mooncake-conductor)
• Reshard (mooncake-reshard)
• Mooncake EP (mooncake-ep)
• Mooncake PG (mooncake-pg)
• Integration (mooncake-integration)
• P2P Store (mooncake-p2p-store)
• Python Wheel (mooncake-wheel)
• Common (mooncake-common)
• Mooncake RL (mooncake-rl)
• CI/CD
• Docs
• Other

Type of Change

• Bug fix
• New feature
• Refactor
• Breaking change
• Documentation update
• Performance improvement
• Other

How Has This Been Tested?

Built the docs locally and checked the rendered conductor pages, the mermaid
diagram, and the mooncake-store-master-publisher anchor. The build produced
no new warnings; the only warnings were intersphinx inventory fetch failures
caused by the local network. Ran pre-commit on the changed files.

Test commands:

cd docs && python -m sphinx -b html source build/html
pre-commit run --files $(git diff --name-only origin/main HEAD)

Test results:

• Unit tests pass
• Integration tests pass (if applicable)
• Manual testing done (describe below)

Every documented field name, default, route, and error code was checked against
mooncake-conductor/src (event_manager.cpp, config.cpp, main.cpp,
event_handler.cpp, zmq_client.cpp, hash_strategy.cpp) and
mooncake-store/src/kv_event/kv_event_publisher.cpp.

Checklist

• I have performed a self-review of my own code
• I have formatted my code using ./scripts/code_format.sh
• I have run pre-commit on the files changed in this PR and all hooks pass
• I have updated the documentation (if applicable)
• I have added tests to prove my changes are effective
• For changes >500 LOC: I have filed an RFC issue

AI Assistance Disclosure

• No AI tools were used
• AI tools were used (specify below)