`wait_for_condition` resets polling state to `initial_state` when a check returns None

[ayushiahjolia]

In wait_for_condition.execute(), state is restored only when checkpointed_result.result is truthy:

if checkpointed_result.is_started_or_ready() and checkpointed_result.result:
    current_state = deserialize(..., checkpointed_result.result, ...)
else:
    current_state = self.config.initial_state

serialize(None) yields None, so a check that returns None gets checkpointed as None, and on resume the falsy check resets current_state to initial_state instead of restoring None. None is overloaded as both "no result yet" and "result is None."

Impact: polling state is lost across attempts for a falsy/None state; a strategy that stops based on state (not attempt) may never complete.

Fix: distinguish a resumed retry from a fresh start via attempt count (not result truthiness) - restore via deserialize when attempt >= 1 (maps None → None), use initial_state only on first start. Add a continue-then-resume test with a None-returning check.

Note: pre-existing on main, surfaced during the #550.

[ayushiahjolia]

None payload resets to initial_state - turned out not a bug, and I did not "fix" it. test_wait_for_condition_retry_without_state explicitly asserts that a checkpointed result=None on a retry uses initial_state (assert result == 6 # 5 (initial) + 1) - that's intended: result=None means "no state was checkpointed," which is distinct from "the state is None." And with the default serdes, a check that returns None is stored as "null" (truthy) and restored correctly, so a real None state is preserved. A falsy payload only comes from a custom serdes violating the -> str contract. My first attempt fought this and broke 3 tests; I reverted it.