perf: Only drop columns and indexes associated with materialized columns if they exist

[tkaemming]

Problem

DROP actions still create mutations, even if those actions don't have anything to drop and these inconsequential mutations are slow to execute. This makes retrying a task unnecessarily slow if after a partial failure or timeout occurs.

Reducing the number of materialized columns we manage is part of #26651.

Changes

Only drop columns and indexes if they actually exist, otherwise skip these steps.

Since these tasks are run via ClickhouseCluster on specific hosts/shards, this should be resilient to partial failures on a per-shard basis.

Does this work well for both Cloud and self-hosted?

N/A, Cloud only

How did you test this code?

Added tests to ensure optimization works as expected, otherwise covered by existing lifecycle tests.