test(pubsub): de-flake test_gossip_gate_filters_peers (#1401)

[yashksaini-coder]

Summary

Closes #1401.

test_gossip_gate_filters_peers fails intermittently on CI (seen on windows (3.12, core); green on all Linux runners and Windows 3.11/3.13):

tests/core/pubsub/test_gossipsub_v1_1_score_gates.py:177
    assert peer1_id in peers_to_send
AssertionError: assert <ID ...> in []

Root cause

The test subscribes all peers, then waits a fixed await trio.sleep(0.2) before calling gsub0._get_peers_to_send(...). That helper skips a topic until it appears in pubsub.peer_topics:

if topic not in self.pubsub.peer_topics:
    continue

peer_topics[topic] is only populated when the node receives a SUBSCRIBE RPC (handle_subscription). On a slow runner the subscription hasn't propagated to gsub0 within 0.2s → topic key absent → continue → send_to empty → peers_to_send == []. A fixed-sleep race, not a logic bug (all mesh/score state in this test is set manually).

Fix

Replace the fixed sleep with the existing event-based helper Pubsub.wait_for_subscription(peer_id, topic) — whose own docstring says "Use this instead of arbitrary trio.sleep() calls to avoid race conditions." Deterministic and non-hanging (5s timeout, returns instantly once the subscription is registered).

-await trio.sleep(0.2)
+await pubsubs[0].wait_for_subscription(hosts[1].get_id(), topic)
+await pubsubs[0].wait_for_subscription(hosts[2].get_id(), topic)

Scope

Intentionally limited to this one test. Sibling tests in the file share the fixed-sleep pattern, but at least one (test_publish_gate_blocks_low_scoring_peers) depends on heartbeat-driven mesh formation, where a subscription-only wait would under-wait and introduce a different flake. A broader de-flake needs per-test analysis and is out of scope here.

Test plan

• test_gossip_gate_filters_peers run 5× locally → stable pass
• Full file test_gossipsub_v1_1_score_gates.py → 11 passed, no regression
• ruff check / ruff format --check clean
• newsfragments/1401.internal.rst added

[acul71]

LGTM — approve.

Correct, well-scoped de-flake for #1401: _get_peers_to_send skips topics absent from peer_topics, and the fixed trio.sleep(0.2) after subscribe raced on slow CI. Switching to Pubsub.wait_for_subscription is the right fix (event-based, 5s timeout, early-return if already registered). Newsfragment and issue linkage look good; CI is green including windows (3.12, core).

Optional follow-up (out of scope, not blocking): a remaining await trio.sleep(0.2) after host connect() in this test (and similar fixed sleeps in sibling score-gate tests) could still flake under connect/mesh races later. Worth a separate audit with per-test readiness predicates when someone has bandwidth — not required for this PR.