fix(simulation): address PR #56 review — pipeline extraction, ActiveDialogue, range check

Tyre #1: Extract run_dialogue_pipeline() shared helper — eliminates ~60
lines of duplication between process_talk_interaction and
process_dialogue_response (L1-L4 pipeline).

Hoshe #1: process_dialogue_response now updates ActiveDialogue with
current tick on follow-up selection — prevents stale started_tick.

Tyre #4: process_dialogue_response now updates InteractionMemory on
follow-up — multi-turn conversations are visible in history.

Hoshe #6 / Tyre #6: handle_dialogue_response adds server-side range
check (CLOSE_RANGE), matching Talk/Confront pattern (D-010 info
boundary).

Hoshe #2: Weighted selection fallback replaced with unreachable!() —
score_line always returns >= 1, so the fallback was dead code.

Hoshe #3: assert!(false, ...) → panic!() in serialization.rs (clippy).

Hoshe #4: SetFacing and TeleportToHub added to roundtrip test.

Hoshe #5: setup_dialogue_response_world inlined (trivial pass-through).

Tyre #2: Doc comment on DialogueCooldownTracker explains per-player-global
design choice (line IDs are NPC-scoped per D-035, no collision risk).

Tyre #3: CONFRONTATION_LINES comment updated with TODO for D-028/D-035
migration.

Tyre #5: DialogueResponse fixture added for cross-language GDScript
testing (input_dialogue_response.msgpack).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

client/tests/fixtures/msgpack/input_dialogue_response.msgpack

@@ -0,0 +1 @@
+うtickﾍ
,ｦaction�DialogueResponseげtarget_entity_id*ｫresponse_idｰkael-davan_d_001