e2e/qa: retry seat withdraw while in-flight rejection persists instead of polling pending flag

[thijsvanemmerik]

Summary

The multicast settlement QA test failed persistently on seat teardown. Two problems in the old flow:

• wait_for_seat_allocation_acked polled the client seat's pending flag to detect the oracle ack. That flag cannot distinguish "the oracle has not acked yet" from "no allocation request was created" (a re-fund of an already-active seat creates none), so the wait could time out with saw_pending=false when there was simply nothing to ack.
• withdraw_seat did a single-shot withdraw. The CLI preflight rejects with an in-flight error while it reads the allocation request PDA as still present, and that read can lag an already-closed request, so a single attempt fails transiently.

Changes:

• Remove the pending-flag ack-wait and make the withdraw retry while the in-flight rejection persists (WithdrawSeatWaitingForAck). The retry absorbs both a slow oracle ack and a lagging RPC read; an in-flight withdraw bails atomically with no state change, so retrying is safe. Any non-in-flight error fails fast.
• Log the seat's tenure and pending flag right after pay (diagnostic only), so a "no request created" run is visible in the run log.

Tradeoff

With the pending-flag ack-wait removed, a genuine oracle outage now surfaces as validate_tunnel_up timing out rather than a specific "not acked" error. The withdraw's timeout error calls out a persistent in-flight rejection (a stale RPC view) versus a slow ack.

Scope

Self-heal of a seat left active by a prior failed teardown is handled separately and is intentionally not part of this PR.

Testing Verification

Verify with a manual workflow_dispatch of qa.mainnet-beta with doublezero_branch set to this PR branch, confirming TestQA_MulticastSettlement passes end to end including withdraw_seat (baseline is green after the stuck seat was cleared).

[thijsvanemmerik]

Closing unmerged. This was a QA-environment issue handled separately; no change to land here.

[ben-dz]

Superseded by #4066