Tolerate refused CLUSTER SLOTS/NODES during topology updates

[mgravell]

Addresses #3274.

Problem

Proxied deployments can identify as a cluster (via HELLO mode, INFO redis_mode, or an answered CLUSTER NODES) while refusing CLUSTER SLOTS with ERR command is not allowed — Redis Enterprise (and so Azure Managed Redis) exposes CLUSTER SLOTS only under the OSS cluster API. The refusal makes await slotsTask in GetEndpointsFromClusterNodes throw RedisServerException, which the outer catch logs as an error ("Encountered error while updating cluster config") — at connect and on every reconfigure, including the periodic topology refresh (default 30 min). The reported stack line (ConnectionMultiplexer.cs:line 2165 in 3.4.0) is exactly that await.

Note the method's own comment already said an error reply for SLOTS should "behave exactly as before" (the pre-4.0 NODES-only path), but the exception fired before that fallback was reached.

An A/B test of the shipped 3.3.1 and 3.4.0 packages against a fake-AMR in-process server showed this path behaves identically in both (same error count at connect and per refresh; only the line number differs, 2162 vs 2165) — so the "works on 3.3.1" aspect of the report is likely about reconfigure frequency (maintenance notifications now auto-enabled for AMR, #3248; dedicated pub/sub connection, #3264) or a server-side change; context requested on the issue.

Fix

A refusal is a posture, not a fault. In GetEndpointsFromClusterNodes, catch RedisServerException per task:

• CLUSTER SLOTS refused → fall back to the CLUSTER NODES-only path, exactly as for a pre-4.0 server;
• CLUSTER NODES refused too → no topology view at all; return null quietly.

Both are logged at Information (new LoggerMessage EventIds 128/129). Genuine faults (timeouts, connection drops, parse failures) still flow to the existing error handling.

Tests

ClusterCommandsRefusedUnitTests ([RunPerProtocol]) drives the in-process test server with a subclass refusing CLUSTER SLOTS (and a variant refusing CLUSTER NODES too, while still announcing itself as a cluster):

• SLOTS refused: endpoint discovery still works from the NODES view alone; no "Encountered error while updating cluster config" is logged; the Information message is.
• Both refused: connect succeeds, PING works, an explicit later ReconfigureAsync stays quiet.

CapturingLogger is promoted from a private nested class in ConnectFailureRefreshTests to a shared test helper.

Full suite: new tests pass across protocols, all *UnitTests green, Build.csproj -c Release /p:CI=true clean.