Blockstore: track when all primary-index data has been purged

[CriesofCarrots]

Problem

#33419 changes the keys of special-column entries; it also adds a fallback so that nodes with mixed data can read both key versions, and checks the old version when the new-version key cannot be found. However, once a node has purged all of its old-format keys, there is no reason to spend these additional db read attempts

Summary of Changes

Track the highest_primary_index_slot in memory
Set this value to None when all the primary-index-key data has been purged
Use this value to optimize various Blockstore read methods, as well as purge_special_columns_exact()

Incidentally, I was originally planning to add to this PR logic to clear the signature-only TransactionMemos in the CompactionFilter when all the primary-index data has been purged. But since the filters are built before the Blockstore element is created, it's a little tricky. Will approach that in a follow-up.

[codecov]

Codecov Report

Merging #33668 (620edcc) into master (21b2fce) will decrease coverage by 0.1%.
Report is 1 commits behind head on master.
The diff coverage is 79.6%.

@@            Coverage Diff            @@
##           master   #33668     +/-   ##
=========================================
- Coverage    81.8%    81.8%   -0.1%     
=========================================
  Files         806      806             
  Lines      217541   217593     +52     
=========================================
+ Hits       178045   178054      +9     
- Misses      39496    39539     +43     

[steviez]

Logic all looks good, just the one nit.

Incidentally, I was originally planning to add to this PR logic to clear the signature-only TransactionMemos in the CompactionFilter when all the primary-index data has been purged. But since the filters are built before the Blockstore element is created, it's a little tricky. Will approach that in a follow-up.

Sounds good!

[steviez]

LGTM