Return merged cache hints from list_all_* client methods #1411

The Peer::list_all_* methods (tools, prompts, resources, resource templates, skills) follow the cursor across pages and return only the items. On 2026-07-28 each page carries ttlMs and cacheScope, and the merged list drops them, so a client that uses list_all_* cannot cache the result correctly.

Proposal: also expose a cache hint for the merged list.

• ttlMs: the earliest expiry across pages. Each page's hint is relative to when that page arrived, so take the minimum of requestTime + ttlMs and convert it back to a TTL relative to the end of the walk.
• cacheScope: public only if every page is public, otherwise private.
• A page without hints (legacy peers, or a server that omits them) makes the merged hint absent.

Changing the return type of the existing methods would be a breaking change, so this would add new API next to them, for example a ListAll<T> { items, ttl_ms, cache_scope } returned by new methods, or a single generic helper.

Suggested by @tsarlandie-oai in #1392 (comment).