[Anish Mehta (anishmehta24)]

Why are these changes needed?

With lexico_objectives, FLOW2.lexico_compare built its history from the first result it got and returned True without comparing it. It is only called once there is an incumbent (and create() starts every local search thread with one), so the incumbent's own result never entered the history and the next result always replaced it, even when it was worse. With one objective and losses 1.0, 5.0, 3.0, 0.5, 2.0, plain FLOW2 keeps the 1.0 incumbent until 0.5 comes in, while lexicographic FLOW2 moves to 5.0 and then 3.0.

The history now starts with the incumbent, and the first result goes through the same comparison as later ones.

Related issue number

None found.

Checks

• I've used pre-commit to lint the changes in this PR (note the same in integrated in our CI checks).
• I've included any doc changes needed for https://microsoft.github.io/FLAML/. See https://microsoft.github.io/FLAML/docs/Contribute#documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

test/tune/test_lexico_flow2.py (min and max mode) fails on main and passes with the change, and pytest test/tune passes. On the Branin-Currin setup from test_lexiflow_performance, the best currin is 2.05 (2.18 on main), still within the test's 2.2 check. Most of the diff in flow2.py is re-indentation from removing the else; git diff -w shows the actual change.