fix increase_parent_counter_cache callback

[kbrock]

Counter caches get corrupted.

It is possible that strategy: sql is the culprit here. There is also probably a problem around cache depth too.

The increase_parent_counter_cache callback is not consistent with the decrease_parent_counter_cache or the udpate

---

some ideas on making them consistent and fixing the problem

• update parent_counter_cache to respects @_trigger_update_callback
• drop @_trigger_delete_callback (are these unnecessary internals?)

Not sure if update follows the same rules as destroy introduced in 63b42f53 But would like both trees to follow the same logic

There are reports of incorrect cache counts. So when insert and destroy have different guard clauses, it is suspect.

I admit that I am not totally familiar to the core use case so I will need to more into why this was introduced in #307 - trying to resolve #584 (and may be able to provide advice for #586 )

TODO:

• introduce tests that corrupt the counter caches see #584 (and mildly related: #586 )
• introduce test with rollback
• introduce STI parents for test (parent should be parent class and child node should be a child class of the parent)

[kbrock]

Feature (not fixing the bug):

• see if we can use REPLACE instead of REGEX_REPLACE. regex buys us anchoring at the beginning of the line (which is needed if we want to support materialized_path), but REPLACE buys us binary support.