Make QueryBuilder.explain_plan() effective and one-shot at execution time

[Copilot]

dsl rough edges identification surfaced a mismatch between DSL ergonomics and runtime behavior: calling .explain_plan() set an internal flag but did not affect execute() unless explain_plan=True was also passed. This PR aligns builder intent with execution semantics.

• Runtime behavior alignment (QueryBuilder.execute)

  • execute() now honors _explain_plan_flag by enabling plan capture for the next call.
  • The flag is consumed after use (one-shot), preventing state leakage across subsequent executions.

• Regression coverage (tests/test_dsl_planner.py)

  • Added a focused test that verifies full lifecycle:
    • .explain_plan() -> first execute() includes result.meta["plan"]
    • second execute() (without re-enabling) does not include plan metadata
    • re-calling .explain_plan() restores plan capture on the same builder instance

q = Q.nodes().compute("degree").explain_plan()
r1 = q.execute(net)                  # includes r1.meta["plan"]
r2 = q.execute(net)                  # no plan metadata
r3 = q.explain_plan().execute(net)   # includes r3.meta["plan"] again