Update sdk-core to main (WFT buffering fix)

changelog/breaking-changes/inimical-beatris.md

@@ -0,0 +1 @@
+Core: Replay detects changes to the service or operation of scheduled Nexus operations, reporting nondeterminism while retaining compatibility with histories recorded before these checks.

changelog/fixed/inimical-beatris.md

@@ -0,0 +1,4 @@
+Core: Worker heartbeats now report correct task-slot and poller counts when using buffered or custom metrics backends, including the Python SDK's `MetricBuffer`. Previously these counts could remain `0` because heartbeat accounting depended on reading labels from backend-owned attributes.
+Core: Local activities still running when their workflow is evicted or the worker shuts down now always receive a cancellation. Local activities no longer retry ordinary failures or start-to-close timeouts after cancellation has been requested, matching non-local activities. Workflow code no longer observes a spurious local activity cancellation when its run is evicted or completes while the local activity is backing off between retries.
+Core: Workers no longer panic when evicting a workflow with buffered local activity results. Completed activations' output is recorded before eviction, without recording eviction-induced cancellations.
+Core: A workflow task that arrives while a workflow waits on a local activity, such as a legacy query, now waits until the current workflow task is reported. Previously it replaced that workflow task ("Trying to send a new WFT for a run which already has one!"), which then timed out on the server and could leave the workflow stuck on a nondeterminism error.