Indigo paged search results

Search stopped at its first page: a query, a profile's posts, a followers list — all returned twenty entries and no way to ask for more (issue #10). This adds paging to every search-kind that has a server cursor.

Changes

• src/app/search.{c,h} — the search model grows a cursor and a has_more flag. indigo_search_begin_page(s, append) stages a fetch (fresh fetch resets, page keeps), indigo_search_finish_page(s, added, next_cursor) completes it, and indigo_search_wants_page is the prefetch gate: near the end of what is held, more are wanted — the same rule the timeline uses, so scrolling does not pause at a page boundary. An empty page on a non-empty list ends the paging even when the server still offers a cursor.
• src/atproto/session.{c,h} — every submit function gains a paging flag; the worker passes the session's cursor to Wolfram only when set, and copies the next cursor into the event before freeing the list. Request size is INDIGO_SEARCH_PAGE (20); the app accumulates up to INDIGO_SEARCH_MAX (60, raised from 20 — the union grew ~100KB and the host test stack limit went with it).
• src/app/app.c — the *_loaded functions append instead of resetting (staging decides), and update_search polls wants_page and maps the search-kind back to its request kind. Saved feeds are the one exception: they come from the account's preferences in a single fetch, so there is no second page to ask for.
• Tests: test_search_paging covers append, cursor stop, and empty-page stop. Existing tests updated for the staging-based reset (a bare loaded call no longer clears). AGENTS.md, README, CHANGELOG updated.

Verification

• make test — 2949 checks, 0 failures
• make warnings — clean
• make snapshots — 111 PNGs, no diffs
• make all with devkitARM — indigo.3dsx built clean

Commit de12fcf, pushed. Issue #10 closed.