Fix Key History Dirty Read Bug

[dillongeorge]

Bug
In the key_history API exposed in directory.rs, it is possible that states read from storage are not fully committed yet (i.e. the transaction has not been committed), but the associated epoch in the states is higher than what is seen in the aZKS entry read from storage. In fact, we explicitly account for that scenario when inspecting versions across the states: https://github.com/facebook/akd/blob/main/akd/src/directory.rs#L484-L490.

Prior to this patch, we always initiated the end_version variable at 0. However, the end_version variable should never be 0 since it will panic in downstream API calls: https://github.com/facebook/akd/blob/main/akd_core/src/utils.rs#L182-L184. The bug in this case is simply that we default end_version to 0, but do not update it in the event that we have performed a dirty read for states.

To resolve the bug, we default end_version to the same value as start_version.

Verification
Prior to updating directory.rs, the added tests to repro cases where a dirty read is performed and a newer epoch is returned as part of the KeyData fails with the expected panic:

failures:

---- tests::test_errors::test_key_history_dirty_reads_experimental_config stdout ----
thread 'tests::test_errors::test_key_history_dirty_reads_experimental_config' panicked at akd_core/src/utils.rs:183:9:
find_max_index_in_skiplist called with input less than smallest element of MARKER_VERSION_SKIPLIST
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

---- tests::test_errors::test_key_history_dirty_reads_whatsapp_v1_config stdout ----
thread 'tests::test_errors::test_key_history_dirty_reads_whatsapp_v1_config' panicked at akd_core/src/utils.rs:183:9:
find_max_index_in_skiplist called with input less than smallest element of MARKER_VERSION_SKIPLIST

failures:
    tests::test_errors::test_key_history_dirty_reads_experimental_config
    tests::test_errors::test_key_history_dirty_reads_whatsapp_v1_config

test result: FAILED. 100 passed; 2 failed; 0 ignored; 0 measured; 0 filtered out; finished in 4.36s

After updating directory.rs, the panic no longer occurs.

[dillongeorge]

Realistically, end_version should never be 0 assuming start_version isn't (i.e. this will short circuit). I'm opting to be a bit more explicit here to err on the side of caution given the bug we're fixing :)

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 77.77778% with 2 lines in your changes missing coverage. Please review.

Project coverage is 88.03%. Comparing base (3ce5335) to head (5818ea1).
Report is 17 commits behind head on main.

Files	Patch %	Lines
akd/src/directory.rs	60.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #449      +/-   ##
==========================================
- Coverage   88.61%   88.03%   -0.58%     
==========================================
  Files          39       38       -1     
  Lines        9109     8283     -826     
==========================================
- Hits         8072     7292     -780     
+ Misses       1037      991      -46     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[kevinlewi]

LGTM!