Looking at WT-6056, I realized that --enable-diagnostic builds are incompatible with the corruption testing we do in tests like test/suite/test_txn19.py and test/csuite/test_wt4156_metadata_salvage.

The problem is the test is written with the expectation the WiredTiger API will return WT_TRY_SALVAGE, or some other error, but in --enable-diagnostic mode builds, failures that panic the system turn into core dumps rather than eventual error returns. Further, it's easy to make a change inside the library that will break a test, just change an error into a panic.

The simplest solution is to turn off --enable-diagnostic for all Evergreen builds. We might want to exempt test/format from the list, and potentially the static test suite (but test_txn19.py should then be skipped in those runs).

cc: luke.chen, alexander.gorrod, sue.loverso