fix(testing): prevent fatal errors from terminating test hosts

Merged
RE
2mo ago
+150-49

Fixes #10409.

A graceful silo shutdown can publish the local membership row as Dead before MembershipTableManager.CurrentStatus is updated. A concurrent refresh then treats the expected shutdown transition as remote eviction and invokes the production fatal handler, terminating the entire shared test process.

This change suppresses fatal handling when the silo lifecycle is already stopping, adds explicit membership convergence barriers to the liveness test, and replaces the production fatal handler in Orleans.TestingHost so unexpected fatal errors stop only the affected silo. It also removes the remaining test-only Environment.FailFast watchdog in fake-time tests in favor of an ordinary asynchronous timeout.

Microsoft Reviewers: Open in CodeFlow