When Connect is called with ServiceName and AbortOnConnectFail=false,
the non-sentinel path returns a disconnected multiplexer that reconnects in
the background. The sentinel path ignores this flag and throws unconditionally
with "Sentinel: The ConnectionMultiplexer is not a Sentinel connection.
Detected as: Standalone".

This is because SentinelConnect returns a disconnected multiplexer
(ServerType defaults to Standalone when the handshake hasn't completed),
and GetSentinelMasterConnection throws when it sees the wrong ServerType
— without checking AbortOnConnectFail.

Changes in GetSentinelMasterConnection:

• Skip the ServerType check when AbortOnConnectFail is false and the
  connection hasn't completed handshake.
• When primary discovery returns null and AbortOnConnectFail is false,
  return a disconnected multiplexer with OnManagedConnectionFailed /
  OnManagedConnectionRestored wired up instead of throwing. Self-healing
  works via the existing SwitchPrimary retry timer.

While Sentinel isn't the future direction for Redis HA, it's still widely
used — especially with Valkey, AWS ElastiCache, and Kubernetes deployments
that haven't moved to cluster mode.

Tests in SentinelAbortOnConnectTests.cs — no running Redis/Sentinel
instance required. Self-healing reuses the existing SwitchPrimary mechanism
tested by ManagedPrimaryConnectionEndToEndWithFailoverTest.