Add ASM (Atomic Slot Migration) cluster maintenance-notification scenario tests

[kiryazovi-redis]

Description of change

Cluster client maintenance-notification scenario tests for the ASM (Atomic Slot Migration) path. All ASM logic lives in the fault injector behind its SLOT_MIGRATE_ASM toggle, so the client stays generic (no ASM-specific code): when the FI runs in ASM mode, the existing effect-trigger cluster tests exercise the FI's reshard trigger and assert SMIGRATING → MaintenanceState.MAINTENANCE → SMIGRATED with node-cache reconciliation.

This PR is therefore small and test-only.

Changes

• Add generic test_notification_handling_with_node_add (+1 node), the mirror of test_notification_handling_with_node_remove.
• Add _wait_for_node_cache_delta settle-poll on the cluster test base: ASM scale (add/remove) is a cascade of SMIGRATING/SMIGRATED (the decision engine moves slots across shards), so the client reconciles its node cache over several notifications after the FI op completes — drain + poll until the ±1 delta before asserting. node_remove now joins the trigger thread + waits for −1 first; reconnect count relaxed to >= 1 (a cascade can mark more than one).
• Skip test_new_connections_receive_last_smigrating_smigrated_notification for all reshard effects (slot-shuffle, add, remove): "a new connection catches the in-flight SMIGRATING" needs a maintenance window that stays open long enough to open a fresh connection mid-flight. ASM add/remove is a multi-window cascade, and a single ASM slot-shuffle over a small slot range closes (SMIGRATED) ~1s after SMIGRATING — the window shuts before a new connection can be opened. Same "ends too fast" class already skipped for slot_shuffle+failover. Existing-connection handling for slot-shuffle stays covered by the handling and cmd-execution tests.
• Size the reshard test timeouts to the FI's real reshard budget so a slow-but-successful reshard is not counted as a failure (raising a timeout only affects the failure path, so it is free on green runs). The FI blocks on a reshard up to 300s (slot-shuffle) / 600s (ASM scale); the client was smaller (op-wait 120s, SMIGRATING waits 60s, SMIGRATED 40s, per-test cap 300s), so a slow reshard flaked and the trigger thread could give up early and leave the command-execution workers hanging. Now: trigger get_operation_result waits RESHARD_OP_TIMEOUT (600s), SLOT_SHUFFLE_TIMEOUT 120→300, SMIGRATED_TIMEOUT 40→120, and the OSS reshard tests use a RESHARD_TEST_TIMEOUT (660s) cap (standalone/non-reshard tests keep 300s).

Requirements

Needs the fault injector with the reshard trigger + SLOT_MIGRATE_ASM toggle. CI runs the generic tests against an ASM-mode fault injector (asm=true).

Pull Request check-list

• Do tests and lints pass with this change? Full single-version (100.0.20) ASM run green — all 4 matrix variants (plain/hiredis × handling/cmd_execution), 0 failures; new_connections[slot-shuffle-reshard] skipped as designed.
• Do the CI tests pass with this change? Validated against an ASM-mode fault injector (asm=true); green end-to-end.
• Is the new or changed code fully tested?

History (condensed)

Earlier revisions added an AsmTriggers enum, include_asm threading, and a dedicated TestClusterClientPushNotificationsWithAsmEffectTrigger with a node-cache "drain until settled" loop. Per review, all ASM logic moved into the FI behind the toggle, so those client-side constructs were removed and the change collapsed to the small, generic diff above. The branch was rebased onto the current feat_add_sch_async base (the earlier huge diff was stale-base divergence, not real change). Follow-up commits then made the generic tests robust to real ASM reshard latency (new-connection skip + FI-budget timeouts) after live ASM CI runs.

---

Note

Low Risk
Test-only changes in scenario integration tests; no production client or library behavior is modified.

Overview
Extends cluster maintenance-notification scenario tests for ASM (reshard) fault-injector runs without changing redis-py client code.

Timeouts are aligned with the FI reshard budget (RESHARD_OP_TIMEOUT 600s, longer SMIGRATED / slot-shuffle waits, RESHARD_TEST_TIMEOUT on OSS reshard tests) so slow-but-successful reshards are not treated as failures and trigger threads do not exit early and leave workers hanging. Non-reshard triggers keep a shorter default op-wait via DEFAULT_TRIGGER_OP_TIMEOUT.

Adds _wait_for_node_cache_delta to drain push notifications and poll until the cluster client’s node cache reaches the expected ±1 change after the FI op completes—needed because ASM scale add/remove is a cascade of SMIGRATING/SMIGRATED events. test_notification_handling_with_node_add mirrors node-remove for a +1 cache delta; node-remove joins the trigger first, waits for −1, and relaxes “marked for reconnect” to >= 1.

Skips test_new_connections_receive_last_smigrating_smigrated_notification when trigger == "reshard" for add/remove/slot-shuffle (maintenance windows too short to open a fresh connection mid-flight). Reshard effect-trigger tests pass RESHARD_OP_TIMEOUT into _trigger_effect.

^{Reviewed by Cursor Bugbot for commit 851a112. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

There are 3 total unresolved issues (including 1 from previous review).

^{Reviewed by Cursor Bugbot for commit 51a8fd6. Configure here.}