sessions: enumerate and inspect prior session transcripts

Problem

The harness exposes limited run-local child transcript inspection, but it has no first-class surface for enumerating persisted prior main-session transcripts and opening one for inspection. Operators who know a conversation occurred cannot discover its session identifier or review its transcript without finding storage implementation details or already possessing an ID.

Proposed capability

Add a first-class session-transcript inventory and inspection surface that lets applicable clients enumerate persisted sessions and retrieve a bounded, paginated transcript for a selected session.

Acceptance criteria

• An operator can list persisted sessions they are authorized to access without already knowing a session ID.
• List results expose stable session identity plus concise metadata useful for selection, such as workspace, timestamps, state, profile/provider/model where available, and a bounded summary or preview.
• An operator can request a selected session transcript with explicit pagination/bounds; the response preserves message/event ordering and distinguishes unavailable or expired content from an empty transcript.
• The surface respects session-store retention, ownership/access boundaries, and redaction rules; it must not expose child/private session content outside its authorized scope.
• Define consistent behavior for stores that cannot efficiently enumerate or retain transcripts, including capability/error semantics rather than silently returning an incomplete inventory.
• Make the capability available through the relevant API/client surfaces and document the persistence and retention expectations.

Relationship

Related to #801. That issue covers search when an operator remembers the topic but not the session. This issue supplies the browse-and-inspect path: enumerate available sessions, select one, and read its transcript. The two capabilities should share identity, authorization, retention, and pagination semantics.

No files or tests are named; start by locating the session-store implementation and the relevant API/client surfaces, then compare their identity, authorization, retention, and pagination behavior with #801. Done means authorized clients can enumerate sessions, retrieve bounded ordered transcripts, receive explicit capability or availability errors, and find the persistence and retention expectations documented.