Pup - the AI's own browser

@@ -11,7 +11,15 @@ legacy.** Tables and reports key on the thread name. Opens already default to th login/...) must accept a missing sessionId and resolve to the caller thread's own window via callerOwnWindow(); then retire sessionId from examples, hints, and describe. Test any verb sweep by calling it with NO sessionId from a thread that owns exactly one window.-ALSO QUEUED: the recovery-binding scenario (open 3 incl. an SPA, restart, binding table 3/3) —+ALSO QUEUED — RECOVERY-BINDING, root cause now FULLY narrowed (2026-07-27, two failed proofs):+all four title/window.name injectors are stamped (1.9.151+1.9.153) and the RESCAN matcher reads+window.name — but the FROM-DISK recovery loop ('Recovering N session(s) from disk', ~line 1160)+binds each session to a page with its OWN logic that never consults window.name; it bound+docs-writer to pages[0] (ESPN) and the real rust-lang page, left unclaimed, was closed as a+leftover. THE ONE REMAINING EDIT: in that loop, before binding, read window.name from candidate+pages and bind ONLY the matching page (fall back to title tag, then lastKnownUrl); never close an+unclaimed page that carries a pup:<sid> name. Prove with: open 3 (one SPA), restart, binding+table 3/3. Prior scenario text: the recovery-binding scenario (open 3 incl. an SPA, restart, binding table 3/3) — the launch-path title injector still needs the window.name stamp (the third injector; two of three were patched in 1.9.151 and docs-writer still mis-bound).