List chats where the current identity is an active participant.

Keyset-paginated, search-aware, and filterable:

• ?archived=true lists chats the caller has archived.
• ?unread_only=true filters to chats with unread messages, excluding effectively-muted chats (TOS-575; resolved in SQL per TOS-634).
• ?q=<text> filters to chats whose title or any active participant's display name matches (case-insensitive substring), across the caller's entire chat history rather than the first page.
• ?cursor=<token> resumes after a previous page. The token for the next page (None on the last page) is returned in the X-Next-Cursor response header so the JSON body stays a plain list.
• ?needs_you_only=true (TOS-7692) keeps only chats whose needs_you is set. It composes with every other filter and with the cursor.

TOS-7692: every row carries last_message_preview, needs_you and working (see services/chats/list_signals.py for their sources), and the X-Needs-You-Count header counts the caller's chats that need them in the requested archive state, ignoring every other filter. Both ride the existing list statements; neither adds a round trip.

TOS-2596 filter axes. Every one of these used to be applied in the browser to whatever page was already loaded, so a filter hid rows instead of running a query - selecting a capture source the caller owned 991 conversations from returned nothing, because none were on page one, and infinite scroll had to be disabled whenever a filter was active. As query params they compose with the cursor, so a page holds limit matches:

• ?source=<slug> restricts to one capture source; the reserved value thinqos selects native conversations (no capture source).
• ?capture_connection_id=<uuid> narrows to one connection/deployment.
• ?capture_org_id=<external org id> narrows to one customer ORG across every connection the caller has for it. Use this, not capture_connection_id, for a user-facing "conversations from <org>" filter: one customer reached through two deployments is two connection rows for the same caller, so a per-connection filter returns part of that customer's history while looking complete.
• ?origin=<slug>&origin=<slug>|<external org id>&origin=thinqos (TOS-4379) selects SEVERAL origins at once, OR'd: a whole source, one customer org of an org-scoped source (slug|org), or native conversations. When present it supersedes source and capture_org_id.
• ?pinned_only=true.
• ?kinds=direct&kinds=room&... OR's the conversation kinds (direct, room, topic, agent, people).
• ?activity_since=<iso8601> bounds by last activity. Send an absolute instant so "Today"/"This week" resolve in the CALLER's timezone.
• ?participant_identity_id=<uuid> narrows to chats that identity is ALSO in (TOS-3292) - an intersection with the caller's own participation, so it can only remove rows. Used by the agent schedule picker to list the chats an agent can be scheduled into.

TOS-2216: the retired ?surface=history split (a separate view of concluded conversations) is gone. concluded_at may still be set internally (see ChatService.conclude_chat/reopen_chat) but no longer excludes a chat from this list - the only user-visible axes are read/unread and archived/unarchived, both participant scoped.