`test_closing_anchorspend_htlc_tx_rbf` still flaky

[nGoline]

#9027 fixed the flake by adding did_short_sig(l1) or in front of the feerate assertions, assuming a short signature was lowering the transaction weight and pushing the feerate above the ±1 ceiling.

However, the test is still failing with the exact same value on this run:

FAILED tests/test_closing.py::test_closing_anchorspend_htlc_tx_rbf - assert (None or 3001.3083296990844 < (3000 + 1))
 +  where None = did_short_sig(<fixtures.LightningNode object at 0x7f1a89bb9c90>)

The identical value 3001.3083296990844 appearing with did_short_sig returning None rules out signature-length variation as the cause. The result is deterministic: fees / weight * 1000 simply rounds to a value just above 3001 for this specific transaction structure, regardless of signature size.

The fix is to replace the hand-rolled ±1 assertions with check_feerate(), which uses a ±2 window and already has the did_short_sig escape hatch built in.