feat(gateway): let players sit, bind and see their intro the way the client expects

[Mathih13]

The 1.12.1 client sends and expects many character, session and world-state messages that LyraCore did not handle. A sit request got "Stand-state changes are not available". A bind gave no " is now your home" line. A new Character never saw its race's intro. Some requests did worse: the protocol library panics on a bad account-data upload and rejects an unknown zone id, and opcodes it cannot read ended the World Session.

This change closes those gaps where the state already exists:

• Stand, sit, sleep and kneel requests reach a new Module verb, gw_set_stand_state. It stores the state in the Character's unit bytes, so observers see it. A ghost can sit, as in vmangos. Damage stands a seated player who is not mounted, as vmangos DealDamage does. The self relay sends SMSG_STANDSTATE_UPDATE on every stored change, and a refused request sends the stored state back so the client snaps back.
• CMSG_EMOTE (the wave after an auction bid) plays its animation through the existing emote event, with no text line.
• A completed innkeeper bind sends SMSG_PLAYERBOUND. It names the Character's zone; vmangos names the sub-area, which the Gateway cannot read without terrain data.
• A first world entry sends SMSG_TRIGGER_CINEMATIC with the race's id from ChrRaces.dbc. Later logins play none.
• World entry and each zone crossing send an empty SMSG_INIT_WORLD_STATES.
• A rename gets SMSG_CHAR_RENAME with CHAR_CREATE_ERROR, because no Character has a rename flag. An account-data request gets an empty section. A second login in one World Session gets SMSG_CHARACTER_LOGIN_FAILED and keeps the session.
• Bug reports, GM surveys, /whois and Warden replies are explicitly ignored. Account-data uploads, zone reports and opcodes the library cannot read are dropped before decoding.

Each packet has a byte-for-byte codec test. Each dispatcher path has a Fake test, and Headless Client tests cover the dropped requests and the cinematic. The durable tests in module/tests/stand_state.rs cover sitting, refusals, a sitting ghost and standing on damage. They were compiled but not run locally; CI's module-durable workflow runs them.

Some vmangos stand-state rules are still missing: auras that sitting or standing cancels, the automatic critical hit on a seated victim, the NOT_STANDING cast failure, 1.5x regeneration while seated, and the UNIT_FLAG_PREVENT_ANIM gate. Account data is not stored, and the bind confirmation names the zone rather than the sub-area. docs/benilla-compatibility.md lists these gaps.

Verification

• cargo fmt --all -- --check, Gateway and Module clippy with -D warnings, cargo test -p lyracore-gateway (2073 passed), Module lib tests (1581 passed), the wasm check, and every durable target with --no-run.
• The durable tests in module/tests/stand_state.rs were not run locally; CI's module-durable workflow runs them.
• Comment review, as CODING_STANDARDS.md defines it. Scope: every added or changed comment in the diff against origin/main, in source, tests and Fakes, plus comments near changed code. Result: passed. The kept comments are one-line contracts (CharacterStore::set_stand_state and stand_state) and technical facts: packet layouts, the ChrRaces.dbc column, the rename result quirk, the first-login read ordering, the reader hazards behind each raw ignore, and the client's stand-state behaviour. The diff removes no comments. I deleted the vmangos citation essays; the behaviour they described is in docs/benilla-compatibility.md.

Made by Claude Opus 5.5 in Claude Code.

🤖 Generated with Claude Code