fix(simplex): make polling fallback reliable

[Zoffka]

Summary

• Adds a dedicated SimpleX polling fallback that seeds seen items, uses per-chat seen keys, and persists them across restarts.
• Keeps the gateway responsive by dispatching polled messages back to the main loop without blocking the poller.
• Marks inbound SimpleX items read from the polling path so sender-visible read receipts are not delayed behind agent work.
• Avoids noisy WebSocket reconnect loops when polling remains healthy.
• Adds an opt-in active-session hard timeout for polling transports so stale turns do not block fresh SimpleX messages indefinitely.

Test plan

• python -m pytest tests/gateway/test_simplex_plugin.py tests/gateway/test_platform_base.py -q -o 'addopts=' → 133 passed
• python -m pytest tests/gateway -q -o 'addopts=' → 5531 passed / 10 failed / 7 skipped; failures are from live environment/config leakage (API server host/model/CORS, Matrix E2EE mock, Signal env), not this SimpleX patch.

[Zoffka]

Closing this PR because it was opened prematurely. I will only reopen/package upstream changes when explicitly requested.

[alt-glitch]

Likely supersedes #27415 (v2) and closed #27360 (v1) — same author, same SimpleX polling fallback work stream. Also overlaps open #26480 (WS reliability) and #26433 (3 SimpleX bugs).

Reviewers: please close #27415 if this PR is the preferred version.