Fix CreateInNotExistAsync bug

[Arkatufus]

Fixes #3397

Problem

AzureApiImpl.ContainerClient() called CreateIfNotExistsAsync() with no exception handling. The Azure SDK has known bugs where this method still throws RequestFailedException(409) for ContainerAlreadyExists. The unhandled exception propagates through the lease actor to the Split Brain Resolver, which reverses its decision to ReverseDownIndirectlyConnected and downs the entire cluster.

Propagation path: ContainerClient() throws 409 → LeaseResourceExists() catch block misidentifies it as a blob-level error → re-thrown as LeaseException → ReadOrCreateLeaseResource() (no catch) → LeaseActor PipeTo → Status.Failure → SBR receives failure → ReverseDownIndirectlyConnected → all nodes downed.

Fix

Replaced CreateIfNotExistsAsync() with CreateAsync() + explicit 409 handling in ContainerClient():

• ContainerAlreadyExists (benign) — silently handled, set _initialized = true
• All other errors — wrapped in a new internal ContainerInitializationException

New ContainerInitializationException — an internal exception type that distinguishes container-level errors from blob-level RequestFailedException. This prevents the existing catch (RequestFailedException) blocks (designed for blob operations) from mishandling container creation errors.

Catch blocks added to LeaseResourceExists(), CreateLeaseResource(), and GetLeaseResource() — each catches ContainerInitializationException and returns its natural "try again" value (false, null, null), which feeds back into the existing retry loop in ReadOrCreateLeaseResource().

Files Changed

File	Change
src/coordination/azure/Akka.Coordination.Azure/Internal/AzureApiImpl.cs	Add ContainerInitializationException; rewrite ContainerClient(); add catch blocks to caller methods
src/coordination/azure/Akka.Coordination.Azure.Tests/AzureApiSpec.cs	Add regression tests for container-already-exists scenario

Test Plan

• ShouldHandleContainerAlreadyExists — reproduces the exact SBR lease-majority fails to acquire lease due to HTTP 409 ContainerAlreadyExists, causing ReverseDownIndirectlyConnected to down the entire cluster #3397 scenario: second AzureApiImpl instance hits pre-existing container, verifies it succeeds instead of throwing
• MultipleInstancesShouldAcquireDifferentLeases — simulates multi-node cluster where two instances race to create the same container
• All 47 existing tests pass (1 pre-existing skip), 0 regressions

[Aaronontheweb]

LGTM