Improve extended buffer pool diagnostics

In the prototype solution rudimentary diagnostics exist in the form of ext_buffer_pool_reads and ext_buffer_pool_pages_flushed counters.

This task is to investigate and improve EBP diagnostics, remove dead or commented out code, resolve TODOs.

It also appears that the counters as written are subject to data races, which should also be investigated and fixed if needed. In any case thread correctness should be documented with comments and enforced with debug assertions (that e.g. the correct lock is held when reading/writing).