Fix LWS surge handling during scale-up and zero-unavailable updates (#778)

* Fix LWS surge handling during scale-up and zero-unavailable updates

Adjust rolling update replica calculation so surge replicas are only reclaimed when permitted by maxUnavailable and scale-up converges directly to the requested replica count.

Add regression tests covering zero-unavailable surge reclamation and scale-up paths that previously created an extra temporary replica.

* Fix surge garbage collection by dynamically calculating maxUnavailable instead of hardcoding subtraction by 1

* Update logging to reflect multiple surge replica deletion

* Add e2e test for MaxSurge + MaxUnavailable

* docs: clarify maxUnavailable and update rollout table for PR 777

* Remove redundant LWS rollout e2e coverage

* Watch leader Pods to complete surge reclamation

---------

Co-authored-by: Adin Ilfeld <adinilfeld@google.com>