Handle blockstore insert dup checks

[carllin]

Problem

should_insert_data_shred() should check just_inserted_data_shreds() for certain shreds

Summary of Changes

Add get_data_shred_from_just_inserted_or_db() to fetch data shreds safely

Fixes #

[sakridge]

lgtm

Pull request has been modified.

[codecov]

Codecov Report

Merging #16051 (2e4e9c7) into master (ff2242d) will increase coverage by 0.0%.
The diff coverage is 100.0%.

@@           Coverage Diff           @@
##           master   #16051   +/-   ##
=======================================
  Coverage    79.9%    79.9%           
=======================================
  Files         409      409           
  Lines      107723   107753   +30     
=======================================
+ Hits        86142    86171   +29     
- Misses      21581    21582    +1     

[carllin]

Ah dang, there's an annoying case here for the new model of handling duplicate blocks as well.

If somebody sends a batch with two shreds both last_index set to true, say for two shred indexes in the order 10, 9. Then the shred_index 9 won't be inserted because it fails the should_insert check. If shred 9 is part of the confirmed version of this slot, then this slot will never finish replay, and moreover, the slot is not marked dead and thus can't be dumped because it's never replayed (the entries are never successfully deserialized blockstore just sees a hole for shred 9 and is waiting for a consecutive set of shreds to deserialize)