More correct alters for non replicated MergeTree

[alesapin]

I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en

Changelog category (leave one):

• Non-significant (changelog entry is not needed)

[alesapin]

Removed this clumsy mutex and corresponding lock, because they are not required (anymore?). Data part cannot be altered from different threads simultaneously because for non replicated MergeTree we execute them sequentially with alter_intention_lock and for ReplicatedMergeTree we use ReplicatedMergeTreeAltreThread executed in schedule_pool which also cannot run concurrently.

UPD: Actually data can be altered concurrently (ALTER PARTITION and ALTER MODIFY COLUMN), but I don't find any cases where alter thread can take columns_lock for read, and then upgrade this lock for write. BTW columns lock will be completely removed with alter on top of mutations.

[alesapin]

test_parts_delete_zookeeper/test.py::test_merge_doesnt_work_without_zookeeper -- flap.