Split session_id from conversation_id

[bgmcmullen]

Problem

conversation_id is overloaded and confuses LLMs (and humans) analyzing the data. Codex
puts a real thread there; Claude has no conversation id so we shove the session id in. But a
Claude session holds the main loop + N subagents, so conversation_id is really a container of
many threads (including subagents, side chats, title generation, suggested input generation and any other api calls in the session)

Model

column	Claude main	Claude subagent	Codex
session_id (partition key, non-null)	session	session	metadata.session_id
conversation_id (nullable)	null	null	thread
agent_id (nullable)	null	agent	null
parent_thread_id (nullable, exists v5)	null	null	parent thread

Touches

• Schema: add session_id, conversation_id nullable, bump version.
• message_projector.js: scope on conversation_id ?? session_id + agent_id.
• Claude: set session_id, null conversation_id. Codex: both + parent_thread_id.
• settle.js: group/load by session_id (only thing that breaks today).
• dataset.js: partition by session_id.

Moves the partition key, so it's breaking — needs a cache recreate + backfill, don't bundle with
additive changes.